Machine Learning of User Attentions in Sensor Data Visualization

Machine Learning of User Attentions in Sensor Data Visualization
Keita Fujino, Sozo Inoue, Tomohiro Shibata Kyushu Institute of Technology

　　Background −　In recent years, along with the spread of smartphones, sensor devices. −　Progress of IoT has resulted in generating an enormous amount of data. −　As a method of utilizing the big data obtained through these devices, visualization is often conducted as the first step. By doing visualization / we can find unusual points or interesting points. This example shows Even looking at this, In general, For example, suppose we focused on this そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points An example of visualization sensor data of a long time The amount of data is large, we don't know where to focus on. ・It is difficult for people to observe sensor data of a long time. 2

　　Background −　If visualize only this attention point, it becomes easy to observe like this. Therefore, At that time, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points It is extremely important to visualize sensor data according to the expertise and purpose of the observer ・Observer never in experienced in data analysis also to be considered. ・Need to know where the observer’s attention points are. 3

Related works − Visualization tools
−　A system that automatically visualizes data has been proposed[1] ・When read data and selects a variable, it automatically visualizes data. −　Development of a tool to represent each layer by holding the original information when zooming in visualized time series data is going on[2] As related research, Visualization tools are proposed. For example, In addition to this, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 5 [1] Kanit Wongsuphasawat, Zening Qu, Dominik Moritz, et al. Voyager 2: Augmenting visual analysis with partial view specications. CHI Conference on Human Factors in Computing Systems (CHI '17), 2017. [2] James S. Walker, Rita Borgo, and Mark W. Jones. Timenotes: A study on effective chart visualization and interaction techniques for time-series data. IEEE Transactions on Visualization and Computer Graphics, 2016. 5

Related works − people's attention points
−　A method of optimize web page design by gathering user’s attention data[3] −　Propose a method to estimate attention points from attention data of people who saw graphic designs and data visualization, and express attention points by a heat map[4] Also, research on people's attention points. For example, In addition, In this way, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points An example of estimated attention points from data visualizations and graphic design[4] −　Various researches on people's attention points are also being done. 6 [3] Xufang Pang, Ying Cao, Rynson W. H. Lau, and Antoni B. Chan. Directing user attention via visual ow on web designs. ACM Transactions on Graphics, 2016. [4] Zoya Bylinskii, Nam Wook Kim, Peter O'Donovan, et al. Learning visual importance for graphic designs and data visualizations. 30tt ACM User Interface Software and Technology Symposium, 2017. 6

Purpose 7 Problems Purpose
−　The bigger the types and amount of sensor data, the more difficult data visualization is. ・Most visualization tools don't consider user's expertise and purpose[3] ・Work efficiency drops, because it takes time to analyze −　It is necessary to collect attention points from each user and each visualization method. However, there are various problems In research on visualization tools, Furthermore, most Also, research on people's attention points, In this research, we aim to At this time, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Purpose Automatic visualization of sensor data using machine learning in consideration of user's expertise and purpose. ・Collect user's attention points ・Apply to other visualization methods through the model trained by a certain visualization method. [3] N. Bikakis, T. Sellis. Exploration and Visualization in the Web of Big Linked Data: A Survey of the State of the Art. LWDM, 2016. 7

Proposed method 8 Precondition Advantage
Sensor data divided by a fixed time length Feature vectors are calculated for each divided section. If more than half of the sections are focused, label T, otherwise label F. The proposed method is like this. The right figure is first, Next, As a precondition, As an advantage, proposed method is そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Precondition Sensor data and attention points the users are collected. Advantage ・Available for various visualization methods ・Can estimate attention points undetectable by the change-point detection method Schematic diagram 8

Proposed method − Training phase points
−　Create a training dataset according to the process. −　Training with this dataset and create an estimation model f. This figure is First, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Training phase of proposed method 9

Proposed method − Estimation phase
−　When there is new sensor data x, it is assumed to be input of the estimation model −　The estimated attention points are outputted. In the estimation phase, As a result, そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Estimation phase of proposed method 10

Data collection 12 ・Collection method of sensor data
−　 An experiment to sense the activities of staff for about three months in nursing care facility with 30 staff members. −　Nursing staff selected from 25 activity labels, including “patrol”, “personal record” and “toilet assistance” from the smartphone app, when carrying out their work. ・Collection method of sensor data 　　Smartphone，Bluetooth LE Sensor device next is behavior(ビヘイビア) sensing at nursing care facilities(ファシリティズ). レジデンツそこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points ・Sensor location 　　Beside each resident's bed，Shared space，Staff’s body 12

　　Evaluation −　We have developed a web app to collect user's attention points. Language：JavaScript Browser：Google Chrome −　We asked user to place a mark with a red frame at the attention point when the user saw the visualization result. Then, next is experimental system そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 13 13

The actual experiment scenery
　　Evaluation −　Display data visualization on the screen and collect user's attention points. −　We collecting users' attention points. This is The actual experiment scenery そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 14 The actual experiment scenery 14

Evaluation 15 15 Outline Subject：5 Males（23 ~ 24 years old, student）
Sensor data：Illuminance，Acceleration Number of data：Illuminance 50 days, Acceleration 10 hours Visualization method：Illuminance Box plot and Bar graph 　　　 Acceleration Plot the three axes X, Y, Z 　　　　　　　　　　　　　 Plot the 3-axis composite value ( Patrol, Personal Record ) Then, next is experiment setting The outline of the experiment is as follows. We extracted the acceleration sensor data / from the time zones labeled(レイベルドゥ) “patrol” and “personal record.” そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 15 Visualization example (Left: Box plot and bar graph, Right: 3 axis X,Y,Z and 3 axis composite value) 15

Evaluation 16 − Number of dataset − Random forest to be used
−　Time window Illuminance ：1 hour Acceleration：2 minutes −　Feature vector Sample size, Maximum, Minimum, Mean, Variance, Standard Deviation, Median, Lower Quartile, Upper Quartile The experimental setting is as follows. we used is the random forest. ディベイションそこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points −　Number of dataset ・Illuminance ・Acceleration (Patrol) ・Acceleration (Personal Record) 6000 1310 1250 −　Random forest to be used 16

Evaluation method 17 Learning Test
1.　Is it possible to estimate the attention points by the proposed method. ・Evaluate from Accuracy using Cross-Validation 2.　Evaluate whether an estimate model of a certain visualization method can be applied to other visualization methods. Learning Test There are ３ evaluation items as follows. When sensor X is present, / visualization A and B are performed, and create estimation models A and B. And at the time of testing / we evaluate whether we can estimate the attention points / by replacing the estimation model. そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Visualization A Estimate model A Estimate model A Visualization A’ Sensor X Sensor X Visualization B Estimate model B Estimate model B Visualization B’ 3.　Compare the proposed method with the change-point detection methods. ・Evaluate using AUC value of the propose method and change-point 　detection(k-nearest neighbor, Singular Spectrum Transformation) 17

Results − Estimation accuracy
− Illuminance sensor and Acceleration sensor labeled “patrol”, an estimation accuracy of 80% or more was obtained. − Acceleration sensor labeled “personal record”, the estimation accuracy was lower than the other two visualization methods. Accuracy for each visualization method そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Sensor Visualization Method Accuracy Illuminance Box plot 85.7% Bar graph 84.5% （Patrol） Line Graph ( X, Y, Z ) 85.0% Line Graph (3-axis composite value) 87.9% （Personal Record） 76.6% 80.8% 18

Results − Apply to visualization method
−　An estimation accuracy of 80% or more for illuminance. −　Decrease in estimation accuracy for Acceleration. Estimation accuracy when the algorithm of another visualization method is applied The "Training" column shows the training visualization method. It shows whether the visualization method can be applied(アプライド) to items in the "test" column / using this estimation model. そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Sensor Training Test Accuracy Illuminance Bar graph Box plot 82.5% 81.8% （Patrol） 3-axis composite value X, Y, Z 70.9% 71.6% （Personal Record） 68.8% 69.9% 19

　 Results　− AUC Value −　In the proposed method, an AUC value of 0.8 or more. −　The change-point detection AUC value is lower than the proposed method. AUC value of proposed method and change point detection Sensor AUC Value Our method K-nn SST Illuminance ( Box plot ) 0.901 0.839 0.683 Illuminance ( Bar graph ) 0.888 0.787 0.659 （Patrol） 0.947 0.556 0.609 （Personal Record） 0.848 0.417 0.427 Next is the result of comparing proposed method and change-point detection そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 20

　　Discussion −　In the case of the illuminance, both of the visualization methods enabled to obtain high accuracy. −　The accuracy of acceleration was lower than that of illuminance. ・The attention points differ depending on subjects. −　Attention points undetectable by the change-point detection method 　 can also be possible to estimate. Then, next is discussion (オブテイン) (アンディテクタブル) そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 21 21

Discussion − Visualization of estimate results
Illuminance −　The place where the amount of change is large is estimated as the attention points. Then, next is visualization of estimate results This example actually visualizing the estimated attention points for illuminance. そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points Example of visualizing estimated attention points (Left: Box plot, Right: Bar graph) 22

Discussion − Visualization of estimate results
Acceleration −　Although there are some gap, it is estimated that almost the same place is attention points. It actually made it easy to see data changes. そこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points 23 Example of visualizing estimated attention points (Left: X, Y, Z, Right: 3-axis composite value) 23

Conclusion 24 Future work
−　We proposed an automatic visualization method of sensor data using machine learning in consideration of user's expertise and purpose. ・Collected attention points of the user who saw the visualization result. ・Calculated the feature vector by dividing the sensor data by a fixed time length, and labeled it with T / F to create training data. −　In the proposed method, accuracy of approximately 80% could be obtained. アプロキシメイトリィそこまで詳しく調べてません I haven’t searched the matter in-depth. It is a future work. 今詳細にお答え出来ません I can’t get into detail I’m not following you, Could you repeat the question more slowly? Q．目的が達成できなくない？（専門性や目的が考慮されてない） A．専門家からのデータを収集する前に、私たちは学生のための実験を行った。この理由は、専門家からのデータを収集して評価する前に、学生との実験をしたかったからです。 Before collecting data from experts, we conducted experiments for students. this reason is, I wanted to do experiments with students before collecting and evaluating data from experts. As a future work, we collect attention points from users with different expertise and purpose. Q. そのデータって何を示しているの？なんでそこなの？（背景のとこ） A. これは例えとして出しているだけなので深い意味はありません。 This is only given as an example so there is no deep meaning. これは行動分析の実験の際に得られた加速度センサです。 This shows the acceleration sensor obtained during the behavior sensing at nursing care facilities Q .結局、着目点のデータを収集しているからRelated workと変わらなくない？ A . 確かに、可視化手法ごと、ユーザごとに着目点データを収集しました。これは評価のための収集であり、1つの注目点データを収集することで実際に複数の視覚化手法に適用できると思います。 Certainly, we collected attention point data for each visualization method and for each user. This is collecting for evaluation, and I think that it can actually be applied to multiple visualization methods by collecting one attention point data. Q.なんでランダムフォレスト？ A. Because I wanted to get the importance of feature vector. And I wanted to evaluate from the importance of this feature vector. Q.なんでその特徴量？ Because it is a basic statistic, we made this a feature vector. Q.実験システムをもっと詳しく If there are multiple attention points, / you can give multiple attention points / by pressing the "Add frame" button. Q. Discussionの図は何？ This shows the points of interest for each subject. Looking at this, in the case of illuminance, focus on the same place, but in the case of acceleration can see that attention points are different. Q. なんで「個人記録」「巡視」なの “Patrol” is a task of visiting the room / of each resident(レジデント) to check abnormalities, / which is an important activity to keep track / of the state of each residentレジデント). / “Personal record” is a task to record information / such as body temperature and blood pressure, physical(フィジカル) condition, and the like. / It is also an important activity to improve(インプルーブ) operational(オペレーショナル) efficiency(エフィシェンスィー), / such as how long it takes to record such information. Q. なんでその可視化なの A bar graph is a general visualization method, and a box plot diagram is a visualization method that can confirm basic statistics. Also, in the case of acceleration, it is common to visualize like this. Therefore, we selected these visualization methods. Q. なんでその時間幅で分割したの？ Because it visualizes illuminance for one day and acceleration about 30 minutes, it was just right if we divide it in 1 hour and 2 minutes. Q. 変化点検出よりも提案手法がいい理由とは？ This is attention points for each user in the proposed method and change point detection. Looking at this, we can estimate the user's attention points with the proposed method, but we can see that it is hardly possible to do with change point detection. The change point score is obtained at each point, and when setting the cut off value so that the accuracy becomes the best it became like this. Q. 今後の展望を詳しく Since I have collected only the attention points from the students this time, I will collect attention points from nursing staff actually in the future. Q. それってなんとなくでつけた着目点じゃない？ When attaching a point of interest, I'm telling "Please mark where you think is important", so I think that it is an important point in each user, that is, attention points ・When applied to other visualization method, the accuracy of illuminance didn’t change but the accuracy of acceleration fell to approximately 70%. ・Attention points undetectable by the change-point detection method can also be estimated. Future work Collect attention points from users with different expertise and purpose. 24

Thank you for listening !!
This is the end of my presentation. do you have any questions?

アプロキシメイトリィ 26

Anomaly detection − k-nearest neighbor
1.Dtestの各要素x(t)において以下を行う A . Dtrainの各要素とx(t)の距離を計算 B. 求めた距離のうち最小のものk個を選び、k近傍から距離を計算し記憶する。 2. 異常度が最大のものを異常部位として列挙 Training Data : Time series Test Data : Window Size : Neighbor : 1. For each element of (a) Calculate the distance between each element of , and (b) The smallest one of the distances obtained above is selected, and the degree 　 of abnormality from the neighborhood of is calculated and stored. 2. Enumerate items with high degree of abnormality as abnormal region. 27

Anomaly detection − Singular Spectrum Transformation
Time Trajectory matrix Test matrix Trajectory matrix 履歴行列 Main subspace 主部分空間 1.履歴行列とテスト行列を作る 2.これらをそれぞれ特異値分解し左差異ベクトルの行列を求める 3.UとQの最大特異値を計算して変化度aを計算する Main subspace Main subspace Discrepancies between both = Variability Perform the following calculation. 1. Make trajectory matrix and test matrix 2. Singular value decomposition of and , and find left singular vectors matrix and 3. Compute the maximum singular value and calculate the variability 28

Evaluation method 29 1-user-image-leave-out cross validation
この項目から実際に膨大なセンサデータを1つの推定アルゴリズムだけで様々な可視化手法に対応して着目範囲を推定して可視化することができるのかを評価する 29

Cross validation（Illuminance）
Subjects：5 Illuminance data：50 User 1, data 1 as test data User 1, data 2 as test data leave-one-out cross-validation User 2, data 1 as test data User 2, data 2 as test data User 5, data 50 as test data 平成29年度　修士論文発表会 30

Comparison of proposed method and change point detection.(illuminance)
近傍法、棒グラフ：AveAccurary: 近傍法、箱ひげ図：AveAccurary: Attention points for each subject k-nn sst 31

Comparison of proposed method and change point detection
Comparison of proposed method and change point detection.(Acceleration) Proposed Method Attention points for each subject k-nn sst Patrol Personal Record 32

Comparison of proposed method and change point detection.(illuminance)
Attention points for each subject k-nn sst 33

Comparison of proposed method and change point detection
Comparison of proposed method and change point detection.(Acceleration) Proposed Method Attention points for each subject k-nn sst Patrol Personal Record 34

　変化度検出の結果 − 照度センサ近傍法特異値スペクトル変換法棒グラフ箱ひげ図平成29年度　修士論文発表会 35

変化度検出の結果 − 加速度センサ 36 近傍法特異値スペクトル変換法加速度（巡視）加速度（個人記録）
　変化度検出の結果 − 加速度センサ近傍法特異値スペクトル変換法加速度（巡視）加速度（個人記録）平成29年度　修士論文発表会 36

ROC curve − Illuminance
K-nn sst Bar graph 左側は箱ひげ図の場合、右側は棒グラフの場合お Box plot 平成29年度　修士論文発表会 37

ROC curve − Acceleration
K-nn sst Acceleration (Patrol) 左側は箱ひげ図の場合、右側は棒グラフの場合お Acceleration (Personal Record) 平成29年度　修士論文発表会 38

Importance of feature Vector − Illuminance
−　"variance" and "standard deviation" is high −　Focus on where data change amount is large 左側はX,Y,Z軸別々の場合、右側は3軸合成値の場合 39

Importance of feature Vector − Illuminance
−　"variance" and "standard deviation" is high −　Focus on where data change amount is large 左側はX,Y,Z軸別々の場合、右側は3軸合成値の場合平成29年度　修士論文発表会 40

Importance of feature Vector
−　Acceleration (Patrol) 左側はX,Y,Z軸別々の場合、右側は3軸合成値の場合平成29年度　修士論文発表会 41

Importance of feature Vector
−　Acceleration (Personal Record) 左側はX,Y,Z軸別々の場合、右側は3軸合成値の場合平成29年度　修士論文発表会 42

Machine Learning of User Attentions in Sensor Data Visualization

Similar presentations

Presentation on theme: "Machine Learning of User Attentions in Sensor Data Visualization"— Presentation transcript:

Similar presentations

About project

フィードバック

ログインする

Auth with social network:

Machine Learning of User Attentions in Sensor Data Visualization

Similar presentations

Presentation on theme: "Machine Learning of User Attentions in Sensor Data Visualization"— Presentation transcript:

Similar presentations

About project

フィードバック