JP2022153048A - Productivity determination apparatus and productivity determination method - Google Patents

Abstract

To provide a productivity determination apparatus capable of easily estimating the productivity of a subject.SOLUTION: A data storage apparatus 4 stores a plurality of pieces of measurement data including time-series data of first biometric information being biometric information on arousal of a task executor in executing a predetermined task, time-series data of second biometric information being biometric information on a higher cognitive function, and time-series data on productivity. A machine learning unit 15 derives an estimation model 14 obtained by learning a relationship between an explanatory variable which is the first and second biometric information included in the measurement data, and an objective variable which is the productivity included in the measurement data, using an algorithm of machine learning. A first measurement apparatus 2 and a second measurement apparatus 3 acquire first biometric information and second biometric information on the subject. A productivity estimation unit 17 estimates the productivity of the subject in a task being executed by the subject, using the derived estimation model 14, based on the acquired first and second biometric information.SELECTED DRAWING: Figure 1

Description

Applied for application of Article 30, Paragraph 2 of the Patent Act Posted date: February 1, 2021 Posted address: https://repository. dl. itc. u-tokyo. ac. jp/? action=pages_view_main&active_action=repository_view_main_item_detail&item_id=55026&item_no=1&page_id=28&block_id=31

The present invention relates to a productivity determination device and a productivity determination method.

Conventionally, for example, an employee's productivity evaluated by a business manager is corrected based on information indicating a decline in employee productivity, such as the employee's physical condition in the past two weeks, and the actual production of the employee is calculated. There is a technique for estimating gender (see, for example, Patent Document 1). In the technique described in Patent Document 1, the employee inputs information indicating the decrease in employee productivity at regular intervals.

WO2014/103248

However, with the technique described in Patent Document 1, since the employee manually inputs information indicating the decrease in the employee's productivity at regular intervals, there is a problem that it takes time and effort to estimate the productivity.
An object of the present invention is to provide a productivity determination device and a productivity determination method that can more easily estimate the productivity of a subject.

In order to solve the above problems, one aspect of the productivity determination apparatus of the present invention includes (a) time-series data of first biometric information, which is biometric information related to the arousal level of the task executor at the time of execution of a predetermined task; (b) a first measurement included in the measurement data; A machine learning unit that derives an estimation model that learns the relationship between the explanatory variables and the objective variables using a machine learning algorithm, with the information and the second biological information as the explanatory variables and the productivity contained in the measured data as the objective variable. , (c) a biological information acquisition unit that acquires the first biological information and the second biological information of the subject, and (d) the first biological information acquired by the biological information acquisition unit using the estimation model derived by the machine learning unit and a productivity estimation unit for estimating the productivity of the subject in the task being executed by the subject based on the second biological information.

Further, another aspect of the productivity determination device of the present invention is to acquire (a) first biological information that is biological information regarding the degree of arousal of the subject, and second biological information that is biological information regarding higher cognitive function. (b) a biometric information acquisition unit, using a task executor's first biometric information and second biometric information at the time of execution of a predetermined task as explanatory variables, and using productivity as an objective variable, using a machine learning algorithm to obtain an explanatory variable and objective variables, based on the first biological information and the second biological information acquired by the biological information acquisition unit, using an estimation model that has learned the relationship between the subject and the objective variable, estimating the productivity of the subject in the task being performed by the subject. and a sex estimator.

In addition, one aspect of the productivity determination method of the present invention is (a) time-series data of the first biometric information, which is biometric information related to the arousal level of the task executor at the time of execution of the predetermined task, related to higher cognitive functions referring to a data storage device storing a large amount of measured data including chronological data of second biological information and chronological data of productivity, and first measured information and second biological information contained in said measured data; Using a machine learning algorithm as an explanatory variable and using the productivity contained in the measurement data as an objective variable, derive an estimation model that has learned the relationship between the explanatory variable and the objective variable, and (b) the subject's Obtaining first biological information and second biological information, and using a derived estimation model based on the obtained first biological information and second biological information, estimating the productivity of the subject in the task being performed by the subject This is the gist of it.

According to one aspect of the present invention, the productivity of a subject can be estimated by inputting the first biological information and the second biological information of the subject. Therefore, it is possible to provide a productivity determination device and a productivity determination method that can more easily estimate the productivity of a subject.

It is a figure showing the whole productivity judging device composition concerning a 1st embodiment. FIG. 4 is a diagram showing measurement data stored in a data storage device; It is a figure which shows the flowchart of a model derivation|leading-out process. It is a figure which shows the flowchart of productivity calculation processing. It is a figure which shows the calculation method of the production amount of the past and the future. FIG. 10 is a diagram showing evaluation results of trained candidate models; FIG. 10A is a diagram showing evaluation results of trained estimation models, where (a) is a diagram showing an estimation model including an electrocardiogram as an explanatory variable and an estimation model not including an electrocardiogram in different colors, and (b) is an explanatory variable; FIG. 10 is a diagram showing an estimated model that includes a pupil diameter as a variable and an estimated model that does not include a pupil diameter in different colors, and FIG. (d) is a color-coded diagram showing an estimated model including skin potentials as an explanatory variable and an estimated model not including skin potentials; (e) is an estimated model including skin temperature as an explanatory variable; It is a figure which shows by color-coding the estimation model which does not include skin temperature. FIG. 10 is a diagram showing evaluation results of a trained estimation model; FIG. 10 is a diagram showing evaluation results of a trained estimation model; FIG. 10 is a diagram showing evaluation results of a trained estimation model; FIG. 10 is a diagram showing a table showing the relationship between the combination of the magnitude of future productivity and the current time period, and the information to be presented to the subject.

An example of a productivity determination device and a productivity determination method according to an embodiment of the present invention will be described below with reference to FIGS. 1 to 11. FIG. Embodiments of the present disclosure will be described in the following order. Note that the present disclosure is not limited to the following examples. Also, the effects described in this specification are examples and are not limited, and other effects may also occur.
1. First Embodiment 1-1. Overall Configuration of Productivity Determining Device 1-2. Model Derivation Processing 1-3. Productivity calculation process 1-4. Estimation model 1-5. Explanatory variables 1-6. Modification 2. Second embodiment

<1. First Embodiment>
[1-1. Overall configuration of productivity determination device]
A productivity determination device 1 according to a first embodiment of the present invention will be described.
FIG. 1 is a diagram showing the overall configuration of a productivity determination device 1 according to the first embodiment. The productivity determination device 1 of FIG. 1 is a device for estimating a subject's productivity (for example, brain labor productivity).
As shown in FIG. 1, the productivity determination device 1 includes a first measuring device 2, a second measuring device 3, a data storage device 4, an input device 5, a display device 6, and a device body 7. ing.

The first measuring device 2 is a device that is attached to the subject's body and that sequentially measures biometric information (hereinafter also referred to as "first biometric information") relating to human arousal. For example, an electrocardiogram (electrocardiographic signal) can be used as the biometric information (first biometric information) relating to human arousal. As an electrocardiogram, for example, a small amount of electric current generated with contraction of the heart can be used. In the first embodiment, an electrocardiogram is used as the first biological information. As the first measuring device 2, for example, an electrocardiograph (ECG) or a smart watch with an electrocardiographic measurement function can be employed. The measured values of the first biological information (electrocardiographic measured values) are sequentially output to the device main body 7 .

The second measuring device 3 is a device that is attached to the subject's body and that sequentially measures biometric information (hereinafter also referred to as "second biometric information") relating to higher human cognitive functions. For example, the pupil diameter can be used as the biometric information (second biometric information) related to higher-order human cognitive functions. In the first embodiment, a case in which the pupil diameter is used as the second biological information will be described. As the second measuring device 3, for example, an eye tracker with a pupil diameter measuring function can be adopted. The measured values of the second biological information (measured values of the pupil diameter) are sequentially output to the device body 7 .

The data storage device 4 is a secondary storage device configured by a HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like. As shown in FIG. 2, the second measuring device 3 measures the time-series data 8 of the electrocardiogram (first biological information) of the task performer and the time-series data 8 of the pupil diameter (second biological information) during execution of the predetermined task. A large number of measurement data 11 including data 9 and time-series data 10 of productivity are stored. As the predetermined task, for example, a mental labor task performed mainly by using the brain and a physical labor task performed mainly by moving the body can be adopted. Moreover, the predetermined task is preferably a task having at least a certain correlation with the task whose productivity is estimated by the productivity determination device 1, and more preferably the same task. As the productivity, for example, the achievement amount of a predetermined task per unit time can be adopted. For example, if the predetermined task is a task of answering a math problem (brainwork task), it may be the number of correct answers per unit time.

Here, it should be noted that the amount of achievement (productivity) of a given task should be carefully defined by analyzing the wrong answer patterns that are likely to occur due to the given task. For example, when the predetermined task is an N-back task, it is more appropriate to use an alignment score than the correct answer rate per unit time for productivity. Experimental data showed that there are two types of error patterns in the N-back task. The two types of incorrect answer patterns are (1) an incorrect answer pattern of no response, and (2) an incorrect answer pattern of doing (N-1)-back task or (N+1)-back task. For example, many subjects who performed the 2-back task incorrectly performed the 1-back task or the 3-back task. The productivity assumed for pattern (1) and pattern (2) is obviously different. In pattern (1) productivity = 0 should be assigned and in pattern (2) productivity higher than 0 should be assigned. If the accuracy rate is used carelessly, pattern (1) and pattern (2) will have the same level of productivity, and if the estimation model learns the data based on it, the relationship between productivity and biometric information will be certain. likely not to learn.

Alignment score is an algorithm that calculates the degree of similarity between two sequences. When pattern (2) occurs, even if the answer is incorrect, a productivity higher than 0 can be assigned as an alignment score as long as the answer sequence and the correct sequence have a certain degree of similarity.

Further, for example, when the predetermined task is RST (Reading Span Task), which is often used in cognitive science, an incorrect answer occurs due to the first impression effect (serial position effect: recency effect). The first impression effect is a phenomenon (early effect, final effect) that makes it easier to remember what you saw first and what you saw last. Therefore, when RST is used, the accuracy rate applied by the quadratic function of the stimulus position can be used as the productivity.

The electrocardiographic time-series data and the pupil diameter time-series data stored in the data storage device 4 are, for example, those measured by the first measuring device 2 and the second measuring device 3 of the productivity determination device 1. Alternatively, it may be measured by another device (for example, an electrocardiograph or an eye tracker with a pupil diameter measurement function different from the first measurement device 2 and the second measurement device 3 of the productivity determination device 1). . Moreover, the time-series data of productivity stored in the data storage device 4 may be measured using, for example, a measurement method according to the task. For example, if the predetermined task is a task of answering a mathematical problem, the computer may be used to measure the questioning of the mathematical problem, the input of the answer, and the measurement of the number of correct answers.

The input device 5 is a device that receives an operation input from a subject. As the input device 5, for example, a keyboard and a mouse can be adopted. An operation input is output to the device body 7 .
The display device 6 is a device for displaying various calculation results and the like by the device main body 7 . As the display device 6, for example, a liquid crystal monitor or a CRT (Cathode Ray Tube) monitor can be adopted.

The device body 7 includes hardware resources such as a storage device 12 and a processor 13 .
The storage device 12 is a secondary storage device configured by an HDD, SSD, or the like. The storage device 12 stores programs executable by the processor 13 . The program is provided by being recorded on removable media, which are packaged media such as magnetic disks, optical disks, magneto-optical disks, and semiconductor memories. Then, the program can be installed in the storage device 12 by attaching the removable medium to the drive. It also stores various data (for example, the estimation model 14) necessary for executing the program.

The processor 13 is various processors such as a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), and an FPGA (Field-Programmable Gate Array). The processor 13 loads a program or the like stored in the storage device 12 into the RAM and executes it, and implements a machine learning unit 15, a biological information acquiring unit 16, a productivity estimating unit 17, a productivity predicting unit 18, and a recommended action presenting unit. 19 and other functions. Then, the machine learning unit 15 executes model derivation processing for deriving the estimation model 14 using a machine learning algorithm based on the large number of measurement data 11 stored in the data storage device 4 . In addition, using the estimation model 14 derived by the biological information acquisition unit 16, the productivity estimation unit 17, the productivity prediction unit 18, and the recommended behavior presentation unit 19, the measured values by the first measurement device 2 and the second measurement device 3 Based on, a productivity calculation process for estimating the productivity of the subject in the task being executed by the subject is executed.

[1-2. Model derivation process]
Next, model derivation processing executed by the machine learning unit 15 will be described. The model derivation process is executed by the subject performing an operation input to the input device 5 for execution of the model derivation process.
As shown in FIG. 3, at step S101, the machine learning unit 15 reads a large number of measurement data 11 stored in the data storage device 4. FIG. Subsequently, preprocessing for machine learning is performed on each of the large number of read measurement data 11 . As the preprocessing, for example, electrocardiogram (first biological information), pupil diameter (second biological information), synchronization of productivity, normalization, and smoothing can be adopted. The contents of the preprocessing are, for example, the type of the estimation model 14 (eg, MLR (Multiple Linear Regression Analysis), SVR (Support Vector Regression), ANN (Artificial Neural Network), CNN (Convolutional Neural Network), LSTM (Long Short-term term Memory))).

Subsequently, the process proceeds to step S102, and after the machine learning unit 15 derives the estimation model 14 based on the measurement data 11 preprocessed in step S101, this model derivation process ends. As a method for deriving the estimation model 14, for example, a method using a machine learning algorithm can be adopted. Specifically, the time-series data 8 of the electrocardiogram (first measurement information) and the time-series data 9 of the pupil diameter (second biological information) included in the measurement data 11 are used as explanatory variables, and the productivity Using the time-series data 10 as the objective variable, a machine learning algorithm is used to derive an estimation model 14 that has learned the relationship between the explanatory variable and the objective variable. As the estimation model 14, for example, a regression model based on a neural network can be adopted. As regression models, for example, MLR, SVR, ANN, CNN, and LSTM can be adopted. In particular, from the point of estimation accuracy, it is preferable to use LSTM as the estimation model 14 . As a result, the machine learning unit 15 obtains the estimation model 14 for estimating the subject's productivity in real time from the measured value of the electrocardiogram (first biological information) and the measured value of the pupil diameter (second biological information).

[1-3. Productivity calculation process]
Next, the productivity calculation processing executed by the biological information acquiring unit 16, the productivity estimating unit 17, the productivity predicting unit 18, and the recommended action presenting unit 19 will be described. In the productivity calculation process, after the first measurement device 2 and the second measurement device 3 are attached to the subject's body, the subject inputs the input device 5 to perform the productivity calculation process. It is executed by performing an operation input for
As shown in FIG. 4, first, in step S201, the biological information acquisition unit 16 obtains the electrocardiogram (first biological information) and the pupil diameter (first 2 biometric information). As a result, the electrocardiogram and the pupil diameter at the time of task execution can be acquired sequentially.

Subsequently, the process proceeds to step S202, and the productivity estimation unit 17 uses the estimation model 14 derived in the model derivation process based on the electrocardiogram and pupil diameter acquired in step S201. to estimate the productivity of Specifically, the productivity output from the estimation model 14 is acquired by inputting the electrocardiogram and the pupil diameter into the estimation model 14 . Thereby, the productivity estimation unit 17 can sequentially acquire the current productivity of the task being executed.

Subsequently, the process proceeds to step S203, and the productivity prediction unit 18 determines whether or not the elapsed time from the start of the productivity calculation process is equal to or longer than a predetermined time Δtp (for example, 5 minutes). If it is determined that the time is less than the predetermined time Δtp (No), the process returns to step S201. On the other hand, if it is determined that the predetermined time Δtp or longer (Yes), the process proceeds to step S204. Thereby, the productivity prediction unit 18 can repeat steps S201 to S203 until the productivity estimation result for the predetermined time Δtp is obtained. Further, after obtaining the result of estimating the productivity for the predetermined time Δtp, the process can proceed to step S204.

In step S204, the productivity prediction unit 18 predicts the future productivity of the subject based on the time-series data of the productivity from the point of time Δtp (5 minutes before) estimated in step S202 to the present. As a method of predicting future productivity, for example, a regression model is used in which the time-series data of productivity from the point in time Δtp (5 minutes ago) to the present is used as an explanatory variable, and future productivity is used as an objective variable. can be used to predict productivity time-series data from the present to the point in time after a predetermined time Δtf (8 minutes). As a regression model, for example, an ARMA (Auto Regressive Moving Average) model can be used.

Subsequently, the process proceeds to step S205, and the recommended action presentation unit 19, as shown in FIG. Based on this, the past production amount Spat is calculated. As a method of calculating the past production amount Spat, for example, a method of time-integrating the productivity from a time point before a predetermined time Δtp (5 minutes before) to the present can be adopted. Subsequently, the recommended action presenting unit 19 calculates the future production volume Sfuture based on the future productivity time-series data for a predetermined time Δtf (8 minutes) from the present predicted in step S203. As a method of calculating the future production amount Sfuture, for example, a method of time-integrating the productivity from the present to the point in time after a predetermined time Δtf (8 minutes) can be adopted.

Next, in step S206, the recommended action presenting unit 19 compares the past production amount Spat calculated in step S204 with the future production amount Sfuture to specify the action that the subject should take. As a method of specifying the action that the subject should take, for example, it is determined whether the production amount per unit time has decreased by 20% or more, and if it is determined that the amount has decreased by 20% or more, the action that the subject should take is "break". and determining that the action to be taken by the subject is "continuation of the task being executed" when it is determined that the decrease has not decreased by 20% or more. Subsequently, after the recommended behavior presenting unit 19 presents the identified behavior to the subject, the process returns to step S201. As a method of presenting the subject, for example, a specific call-to-action message (e.g., "Production will decrease at this rate. Please take a break." "No need to take a break. Continue your current job. ”) on the display device 6 can be adopted.

As a result, in the productivity calculation process, by repeating the flow of steps S201 to S206, the comparison between the past production amount Spat and the future production amount Sfuture is repeated, and it is predicted that the subject's production amount will decrease. , a message prompting you to take a break is displayed. Therefore, the subject can be given a break before production actually decreases, and the subject's productivity can be restored. Therefore, although the production amount during the break is "0", the production amount can be improved as a whole.

As described above, in the productivity determination device 1 according to the present embodiment, the machine learning unit 15 obtains biological information (first measurement information) related to the arousal level of the task executor and high-order Using biometric information (second biometric information) related to cognitive function as an explanatory variable and productivity as an objective variable, a machine learning algorithm is used to derive an estimation model 14 that has learned the relationship between the explanatory variable and the objective variable. Then, the productivity estimating unit 17 uses the estimation model 14 derived by the machine learning unit 15, based on the first biological information and the second biological information of the subject, in the task being executed by the subject, estimating the productivity of the subject. presume. Therefore, by inputting the first biological information and the second biological information of the subject, the productivity of the subject can be estimated. Therefore, it is possible to provide the productivity determination device 1 capable of estimating the productivity of the subject more easily.

Further, in the productivity determination device 1 according to the present embodiment, the recommended action presenting unit 19 calculates the past productivity based on the time-series data of the productivity from the point in time before the predetermined time Δtp estimated by the productivity estimating unit 17 to the present. In addition to calculating the production amount Spat, the future production amount Sfuture is calculated based on the future productivity predicted by the productivity prediction unit 18 until a predetermined time Δtf from the present, and the calculated past production amount Spat and the future production amount The action to be taken by the subject is specified by comparing with Sfuture, and the specified action is presented to the subject. Thereby, for example, when a decrease in the production of the subject is predicted, the subject can be encouraged to take a break before the actual decrease in the production.

[1-4. estimation model]
Next, a method for selecting the estimation model 14 will be described. Selection of the estimation model 14 is performed by evaluating the trained candidate models from two aspects: evaluation of estimation performance using statistical indicators and evaluation of reliability using interpretation techniques.
Regarding Evaluation of Estimation Performance Using Statistical Index First, a plurality of types of candidates for the estimation model 14 (hereinafter also referred to as “candidate models”) are prepared.
Candidate models include, for example, MLR, SVR, ANN, CNN and LSTM. Subsequently, for each prepared candidate model, a machine learning algorithm is used to derive a trained candidate model that has learned the relationship between the explanatory variable (first biological information, second biological information) and the objective variable (productivity). , perform performance evaluation of the derived trained candidate models. As the performance evaluation, for example, an evaluation index of the error between the estimated productivity value obtained from the trained candidate model and the measured productivity value (hereinafter also referred to as "first evaluation index"), the trained candidate model It is possible to adopt an evaluation index (hereinafter also referred to as a second evaluation index) of the degree of matching between the change trend of the estimated productivity value obtained from and the change trend of the measured productivity value. As the estimated value of productivity, for example, the time-series data 8 of the electrocardiogram (first measurement information) included in the measurement data 11 not used for learning the candidate model, and the time-series data 8 of the pupil diameter (second biological information) By inputting 9 into the trained candidate model, the productivity output from the trained candidate model can be adopted. As the measured value of productivity, for example, time-series data 10 of productivity included in the measurement data 11 can be used. For example, RMSE (Root Mean Square Error) can be used as the first evaluation index. As the second evaluation index, for example, trend accuracy, DTW (Dynamic Time Warping), cosine similarity, and R (Pearson's correlation coefficient) can be used.
RMSE is the square root of the residual variance and is calculated by the following equation (1). In the following equation (1), pk (k=1, 2, 3, . . . , N) is the measured value of productivity, and pk is the estimated value of productivity. The RMSE becomes smaller as pk and pk are closer (as the estimation accuracy is higher).

Ｒは、下記（３）式で算出される。下記（３）式では、ｐ^－はｐkの相加平均、ｐ^^－はｐ^の相加平均である。Ｒは、ｐkとｐ^kとの動きのトレンドが近づくほど大きくなる。特に、Ｒは、データの小さいズレに対するロバスト性や、ノイズに対するロバスト性の点から、Trend Accuracy、ＤＴＷ、コサイン類似度よりも、第２評価指標として好ましい。

Trend Accuracy is calculated by the following formula (2). In the following equation (2), σk is "1" if the sign of (dp/dt)k and (dp^/dt)k match, and is "0" if they do not match. Trend Accuracy increases as the trend of movement between pk and pk approaches.

R is calculated by the following formula (3). In the following equation (3), p- is the arithmetic mean of pk, ^{and p^- is} the arithmetic mean of p^. R increases as the trend of movement between pk and pk approaches. In particular, R is more preferable than Trend Accuracy, DTW, and cosine similarity as the second evaluation index in terms of robustness against small deviations in data and robustness against noise.

Subsequently, based on the calculated first evaluation index and second evaluation index, select the best candidate model from among multiple types of candidate models (MLR, SVR, ANN, CNN, LSTM), and select the selected candidate model It is used as the estimation model 14 . FIG. 6 is a diagram showing evaluation results of trained candidate models. As a result of learning and evaluating the candidate models, as shown in FIG. 6, the first evaluation index and the second evaluation index showed the best values when the LSTM was used. FIG. 6 illustrates a case where RMSETrend Accuracy is used as the first evaluation index and the second evaluation index. Although illustration is omitted, similar results were obtained when R was used as the second index value.

Evaluation of Reliability Using Interpretation Technique First, candidate models whose first index value and second index value satisfy threshold values are extracted in the above estimation performance evaluation using statistical indices. For example, extract LSTM, SVR, and MLR. Next, an interpretation diagram is created for the extracted candidate models using ALE (accumulated effect plot). As an interpretation diagram, for example, a graph is created in which the horizontal axis is the feature amount (explanatory variable) and the vertical axis is the output of the candidate model (objective variable). ALE is an evaluation technique that can be used even when the features are not independent of each other. Next, using the created interpretation diagram, the relationships between the feature values and the outputs of the candidate models are identified from among the candidate models. ) is selected as a candidate model with high confidence. As a result of evaluating the reliability, the reliability showed the best value when LSTM was used. Therefore, as a result of the evaluation of the above two aspects, when MLR, SVR, ANN, CNN and LSTM are used as candidate models, the estimation model has superior estimation performance and high reliability LSTM was found to be optimal.

[1-5. explanatory variable]
Next, the validity of using electrocardiogram and pupil diameter as explanatory variables will be described.
First, we prepared multiple candidate explanatory variables. As candidates, for example, electrocardiogram, pupil diameter, respiration, skin potential, and skin temperature alone or in combination were adopted. Subsequently, for each candidate prepared, an estimation model 14 was derived using the candidate as an explanatory variable, and performance evaluation of the derived estimation model 14 was performed. 7(a), 7(b), 7(c), 7(d), and 7(e). It is plotted on a graph in which the axis is the first evaluation index and the vertical axis is the second evaluation index. In FIG. 7A, the color of the circle is changed depending on whether or not the electrocardiogram is included in the explanatory variables. Similarly, in FIG. 7B, whether or not the explanatory variable includes the pupil diameter, in FIG. 7C, whether or not the explanatory variable includes respiration, and in FIG. No, in FIG. (e), the color of the circle is changed depending on whether the pupil diameter is included in the explanatory variable. FIGS. 7A to 7E illustrate a case where RMSE and R are used as the first evaluation index and the second evaluation index, and LSTM is used as the estimation model 14. FIG. The same applies to FIGS. 8, 9 and 10 as well. Although illustration is omitted, similar results were obtained when R was used as the second index value.

As a result, in the graph shown in FIG. 7A, the estimated model 14 including the electrocardiogram as explanatory variables (hereinafter also referred to as the "model with electrocardiogram 14") and the estimated model 14 not including the electrocardiogram (hereinafter referred to as Clustering (grouping) was possible with "model 14 without electrocardiogram"). Also, the group to which the model 14 with electrocardiogram belongs showed better values in the first evaluation index and the second evaluation index than the group to which the model 14 without electrocardiogram belongs. Therefore, in terms of estimation accuracy, it can be said that the estimation model 14 in which the electrocardiogram is included as an explanatory variable is preferable. On the other hand, in the graph shown in FIG. 7B, the estimation model 14 including the pupil diameter as an explanatory variable (hereinafter also referred to as "model 14 with pupil diameter") and the estimation model 14 not including pupil diameter are clustered ( Grouping) was possible, but the group to which the model 14 with pupil diameter belongs did not necessarily show good values in the first evaluation index and the second evaluation index. In addition, the estimation models 14 could not be clustered in the graphs shown in FIGS. 7(c) to 7(e).

7(a) to 7(e) are replaced with figures (●+▲■ ★) was converted to a graph. In the graph shown in FIG. 8, when the number of types of biometric information is "2" or more, the first evaluation index and the second evaluation index show better values than when the number is "1". Moreover, as shown in FIG. 9, a graph was created by grouping the figures shown in FIG. 8 by the types of biological information included in the explanatory variables of the estimation model 14. The groups include a group of estimation models 14 that include both electrocardiogram and pupil diameter as explanatory variables (hereinafter also referred to as “first group”), and a group of estimation models 14 that include electrocardiography and do not include pupil diameter as explanatory variables. group (hereinafter also referred to as "second group"), the group of the estimation model 14 that includes the pupil diameter as an explanatory variable and does not include the pupil diameter (hereinafter also referred to as "third group"), the electrocardiogram and the pupil diameter as explanatory variables A group of estimation models 14 (hereinafter also referred to as a "fourth group") that does not include the . In the graph shown in FIG. 9, the first group, second group, third group and fourth group could be clustered without overlapping each other. Then, the first evaluation index and the second evaluation index showed favorable values in the order of the first group>second group>third group>fourth group. Therefore, according to FIGS. 8 and 9, from the point of view of estimation accuracy, the number of types of biological information included in explanatory variables is preferably two or more. It can be said that preferably the diameter is included.

Further, the performance evaluation results of the estimation model 14 shown in FIGS. A graph was created in which the indicators are arranged in order of size. In the graphs of FIGS. 10(a) and 10(b), when the estimation model 14 including all of the electrocardiogram, pupil diameter, respiration, skin potential and skin temperature as explanatory variables is used, the second evaluation index is the highest. showed good value. Therefore, it can be said that the estimation model 14 that includes everything from the electrocardiogram to the skin temperature as explanatory variables is preferable from the point of estimation accuracy using the second evaluation index. On the other hand, when using the estimation model 14 that includes only two explanatory variables, the electrocardiogram and the pupil diameter, the second evaluation index was the third best value, but the first evaluation index was the best. showed good value. Therefore, the estimation model 14 containing only two of the electrocardiogram and pupil diameter as explanatory variables is inferior to the estimation model 14 containing all of the electrocardiogram to skin temperature as explanatory variables in the second evaluation index, but the first evaluation It is the best index, and since the amount of biometric information is small, there is little effort required to measure biometric information, and the amount of computation required for estimating productivity is also small. Therefore, it can be said that it is most preferable to use a combination of the electrocardiogram and the pupil diameter as explanatory variables in terms of the estimation accuracy, the labor of measurement, and the amount of calculation.

In addition, using the combination of the electrocardiogram and the pupil diameter as an explanatory variable for the estimation model 14 is also appropriate from the viewpoint of factors that reduce productivity. In other words, one of the main reasons for the decrease in productivity is the decrease in alertness. In addition, the decline in productivity is also related to the decline in higher cognitive functions such as concentration and information processing ability. On the other hand, the electrocardiogram is biological information related to arousal. In addition, the pupil diameter is biological information related to higher-order cognitive functions. Therefore, it can be said that it is preferable to use the combination of the electrocardiogram and the pupil diameter as an explanatory variable also from the viewpoint of the factor of the decrease in productivity.

[1-6. Modification]
(1) In the first embodiment, the recommended action presentation unit 19 compares the past production amount Spat and the future production amount Sfuture to specify and present an action that the subject should take. However, other configurations can also be employed. For example, it may be configured to identify and present actions that the subject should take based on future productivity. In this case, for example, when it is determined that the future productivity is equal to or less than a predetermined threshold, by presenting a message prompting a break, calculation of the past production amount Spat and the future production amount Sfuture can be omitted, Amount of calculation can be reduced.
(2) Also, for example, as shown in FIG. 11, referring to a table showing the relationship between the combination of the magnitude of future productivity and the current time period, and the information to be presented to the subject, actions to be taken by the subject may be configured to specify. In the table shown in FIG. 11, future productivity is divided into three categories of "high", "medium", and "low", and time zones are divided into four categories of "morning", "noon", "evening" and "midnight". Predetermined information (for example, "today's schedule") is written for each combination.

(3) In addition, in the first embodiment, an example in which both the model derivation process and the productivity calculation process are executed by one productivity determination device 1 is shown, but other configurations can also be adopted. . For example, it may be executed by different productivity determination apparatuses 1 from each other. In this case, one productivity determination device 1 executes model derivation processing, performs machine learning to derive an estimated model 14, and stores the derived estimated model 14 in the storage device 12 of the other productivity determination device 1. By storing, in another productivity determination device 1, it is possible to estimate the subject's productivity using the estimation model 14 stored in the storage device 12, omit the model derivation process, and perform the productivity calculation process can be executed immediately.

<2. Second Embodiment>
Next, the productivity determination device 1 according to the second embodiment will be described. The overall configuration of the productivity determination device 1 according to the second embodiment is the same as that in FIG. 1, so the illustration is omitted.
The productivity determination device 1 according to the second embodiment differs from the first embodiment in the explanatory variables of the estimation model 14 . In the second embodiment, the machine learning unit 15, in addition to the electrocardiogram (first biological information) and the pupil diameter (second biological information), third biological information other than the first biological information and the second biological information, Using the fourth biological information and the fifth biological information as explanatory variables and productivity as the objective variable, an estimation model 14 that has learned the relationship between the explanatory variables and the objective variable is derived using a machine learning algorithm. For example, respiration, skin potential, and skin temperature can be used as the third, fourth, and fifth biological information. In addition, the biological information acquisition unit 16, electrocardiogram (first biological information), pupil diameter (second biological information), respiration (third biological information), skin potential (fourth biological information) and skin temperature (fifth biological information information). Further, the productivity estimation unit 17 uses the estimation model 14 derived by the machine learning unit 15 based on the acquired first biological information, second biological information, third biological information, fourth biological information, and fifth biological information. to estimate the subject's productivity in the task the subject is performing.

Thus, in the productivity determination device 1 according to the second embodiment, the estimation model 14 including the first biological information, the second biological information, the third biological information, the fourth biological information, and the fifth biological information as explanatory variables , for example, compared to the case of using the estimation model 14 that includes only the first biological information (electrocardiogram) and the second biological information (pupil diameter) as explanatory variables, the accuracy of estimating the productivity of the subject can be improved ( See FIG. 10(b)). In FIG. 10(b), R of the estimation model 14 including all of electrocardiogram, pupil diameter, respiration, skin potential and skin temperature as explanatory variables has the best value.

As described above, according to each aspect of the present invention, by measuring and inputting the first biological information and the second biological information of the subject by the first measuring device 2 and the second measuring device 3, the productivity The actual productivity of the subject can be easily estimated without the subject manually inputting information used for estimating . In addition, in the estimation, learning is performed using measurement data 11 including past productivity in addition to biological information (first biological information) regarding arousal level and biological information (second biological information) regarding higher cognitive function. Since the estimation model 14 is used, the actual productivity of the subject can be estimated more accurately. Furthermore, as the estimation model 14, it was found from long-term research that LSTM (Long Short-Term Memory) was the most suitable among many models, and this was applied. This made it possible to further improve the estimation accuracy. Furthermore, in each aspect of the present invention, in addition to simply estimating productivity, future productivity is predicted from information including changes in productivity, and based on the predicted future productivity, what is the subject currently doing? Since the recommended action presenting unit 19 that points out whether the action should be taken is further provided, it is possible to prevent a decrease in productivity in advance.

1... productivity determination device, 2... first measuring device, 3... second measuring device, 4... data storage device, 5... input device, 6... display device, 7... device body, 8... electrocardiogram (first measurement information) time-series data, 9 time-series data of pupil diameter (second biological information), 10 time-series data of productivity, 11 measurement data, 12 storage device, 13 processor, 14 estimation model, 15... Machine learning unit, 16... Biological information acquisition unit, 17... Productivity estimation unit, 18... Productivity prediction unit, 19... Recommended action presentation unit

Claims (10)

Time-series data of the first biometric information, which is biometric information regarding the arousal level of the task executor, time-series data of the second biometric information, which is biometric information regarding higher cognitive function, and time of productivity when executing the predetermined task a data storage device storing a large number of measured data including series data;
Using the first measurement information and the second biological information contained in the measurement data as explanatory variables and the productivity contained in the measurement data as the objective variable, a machine learning algorithm is used to determine the relationship between the explanatory variables and the objective variable. a machine learning unit that derives a learned estimation model;
a biological information acquisition unit that acquires the first biological information and the second biological information of a subject;
Using the estimation model derived by the machine learning unit, based on the first biological information and the second biological information acquired by the biological information acquisition unit, estimate the productivity of the subject in the task being performed by the subject and a productivity estimating unit. the first biological information is an electrocardiogram;
The productivity determination device according to claim 1, wherein the second biological information is a pupil diameter. The productivity determination device according to claim 2, wherein the explanatory variables are only an electrocardiogram as the first biological information and a pupil diameter as the second biological information. The productivity determination device according to any one of claims 1 to 3, wherein the estimation model is an LSTM (Long Short-Term Memory). a productivity prediction unit that predicts the future productivity of a subject based on the productivity estimated by the productivity estimation unit from a time point a predetermined time ago to the present;
5. The productivity determination device according to any one of claims 1 to 4, further comprising a recommended action presenting unit that specifies and presents actions to be taken by the subject based on the future productivity predicted by the productivity predicting unit. . The recommended action presentation unit calculates the past production volume based on the time-series data of the productivity from the point in time before the predetermined time estimated by the productivity estimation unit to the present, and the current production predicted by the productivity prediction unit. Calculate the future production volume based on the time-series data of the future productivity from the time to a predetermined time later, compare the calculated past production volume and the future production volume, and identify the action to be taken by the subject , presents the identified behavior to the subject. A biological information acquisition unit that acquires first biological information that is biological information related to arousal of the subject and second biological information that is biological information related to higher cognitive functions;
Using the first biological information and the second biological information of the task executor at the time of execution of a predetermined task as explanatory variables and productivity as an objective variable, a machine learning algorithm is used to calculate the relationship between the explanatory variables and the objective variables. Productivity estimation for estimating the productivity of a subject in a task being performed by the subject based on the first biological information and the second biological information acquired by the biological information acquisition unit using an estimation model that has learned the relationship and a productivity determination device. Time-series data of the first biometric information, which is biometric information regarding the arousal level of the task executor, time-series data of the second biometric information, which is biometric information regarding higher cognitive function, and time of productivity when executing the predetermined task Referring to a data storage device storing a large number of measurement data including series data, using first biological information and second biological information contained in the measurement data as explanatory variables, and using productivity contained in the measurement data as an objective variable, Using a machine learning algorithm, derive an estimation model that has learned the relationship between the explanatory variable and the objective variable,
Obtaining the first biological information and the second biological information of the subject, and using an estimation model derived based on the obtained first biological information and the second biological information, the subject in a task being executed by the subject A productivity determination method for estimating the productivity of A plurality of types of candidate models that are candidates for the estimation model are prepared, and for each of the prepared candidate models, a machine learning algorithm is used to derive a candidate model that has learned the relationship between the explanatory variable and the objective variable, and the derived learning. A first evaluation index that is an evaluation index of the error between the estimated productivity value and the measured productivity value obtained from the trained candidate model, and the change trend of the estimated productivity value obtained from the trained candidate model Calculate a second evaluation index that is an evaluation index of the degree of matching with the change trend of the measured value of productivity, and based on the calculated first evaluation index and the evaluation index, select the best candidate model from among multiple types of candidate models. and using the selected candidate model as the estimation model. The first evaluation index is RMSE (Root Mean Square Error),
The productivity determination method according to claim 9, wherein the second evaluation index is R (Pearson's correlation coefficient).

Images