Prediction system, prediction method, and program

Abstract

The present invention accurately predicts the magnitude of a subject's future psychological stress. A prediction system (100) comprises: a first extraction unit (42) that extracts a group of similar workers from a plurality of workers (W) on the basis of the degree of similarity of the work patterns of the plurality of workers (W) to the work pattern of a subject (O) during a first period; a first model generation unit (43) that generates a first prediction model (57) through machine learning using, as learning data, first information including at least work pattern information, measurement information, behavioral record information, and psychological stress information of each worker (W) included in the group of similar workers; and a first prediction unit (44) that predicts the magnitude of psychological stress of the subject (O) by entering input data, including at least work pattern information, measurement information, and behavioral record information of the subject (O), into the first prediction model (57).

Description

Prediction system, prediction method, and program

[0001] The present invention relates to a prediction system for predicting the level of psychological stress of a subject.

[0002] Depression is a mental illness with a high prevalence and a high recurrence rate. Therefore, depression tends to last for a long time, resulting in economic losses to society. For this reason, there is an urgent need to prevent depression. One way to prevent depression is to recognize mental health problems at the pre-symptomatic stage and take appropriate measures. For example, Patent Document 1 discloses a technology for preventing and improving mental illness by automatically calculating the risk of mental health problems and informing the individual, medical professionals, managers, etc. of the results and methods for prevention and improvement.

[0003] Japanese Patent Application Publication No. 2022-176775

[0004] In order to more effectively prevent depression, it is important to accurately predict the future state of mental health, that is, the magnitude of future psychological stress.

[0005] One aspect of the present invention aims to realize a prediction system or the like that can accurately predict the degree of future psychological stress of a subject.

[0006] In order to solve the above problem, a prediction system according to one embodiment of the present invention comprises: a first extraction unit that extracts a group of similar workers from a plurality of workers based on the similarity of their working styles with the working style of a subject person during a first period; a first model generation unit that generates a first prediction model by machine learning using, as learning data, first information that includes at least: working style information that indicates the working style of each worker included in the group of similar workers; measurement information that includes the activity level and sleep duration of each worker; behavior record information that records the time of behavior performed by each worker by behavior type; and psychological stress information that indicates the level of psychological stress of each worker; and a first prediction unit that predicts the level of psychological stress of the subject person from the first period onwards by inputting input data that includes at least the working style information, the measurement information, and the behavior record information of the subject person into the first prediction model.

[0007] In order to solve the above problem, a prediction system according to one embodiment of the present invention comprises: a second extraction unit that extracts a group of similar workers from a plurality of workers based on the degree of similarity of their attendance status with the attendance status of a subject person during a first period; a second model generation unit that generates a second prediction model by machine learning using, as learning data, second information that includes at least: attendance status information that indicates the attendance status of each worker included in the group of similar workers; measurement information that includes the activity level and sleep time of each worker; behavior record information that records the time of behavior performed by each worker by behavior type; and psychological stress information that indicates the level of psychological stress of each worker; and a second prediction unit that predicts the level of psychological stress of the subject person from the first period onwards by inputting input data that includes at least the attendance status information, the measurement information, and the behavior record information of the subject person into the second prediction model.

[0008] In order to solve the above problem, a prediction method according to one aspect of the present invention includes: a first extraction step of extracting a group of similar workers from a plurality of workers based on the degree of similarity of their work styles with the work style of a subject person during a first period; a first model generation step of generating a first prediction model by machine learning using, as learning data, first information including at least: work style information indicating the work style of each worker included in the group of similar workers; measurement information including the activity level and sleep time of each worker; behavior record information that records the time of behavior performed by each worker by behavior type; and psychological stress information indicating the level of psychological stress of each worker; and a first prediction step of predicting the level of psychological stress of the subject person from the first period onwards by inputting input data including at least the work style information, the measurement information, and the behavior record information of the subject person into the first prediction model.

[0009] In order to solve the above-mentioned problems, a prediction method according to one aspect of the present invention includes: a second extraction step of extracting a group of similar workers from a plurality of workers based on the degree of similarity of their attendance status with the attendance status of a subject person in a first period; a second model generation step of generating a second prediction model by machine learning using, as learning data, second information including at least: attendance status information indicating the attendance status of each worker included in the group of similar workers; measurement information including the activity level and sleep time of each worker; behavior record information recording the time of behavior performed by each worker by behavior type; and psychological stress information indicating the level of psychological stress of each worker; and a second prediction step of predicting the level of psychological stress of the subject person from the first period onwards by inputting input data including at least the attendance status information, the measurement information, and the behavior record information of the subject person into the second prediction model.

[0010] The prediction system in each aspect of the present invention may be realized by a computer. In this case, the control program for the prediction system that causes the computer to operate as each part (software element) of the prediction system to realize the prediction system on a computer, and the computer-readable recording medium on which the control program is recorded, also fall within the scope of the present invention.

[0011] According to one aspect of the present invention, it is possible to accurately predict the degree of future psychological stress of a subject.

[0012] 1 is a block diagram showing an example of the configuration of a prediction system according to a first embodiment of the present invention. FIG. 2 is a flowchart showing an example of the flow of processing performed by a server according to the first embodiment of the present invention. FIG. 3 is a diagram showing an ROC curve created based on prediction results when a conventional prediction model is used and when a first prediction model generated by the method described in this embodiment is used, and the actual K6 score of each worker one week after the prediction is made. FIG. 4 is a graph showing the relationship between the rate of teleworking and the level of psychological stress. FIG. 5 is a block diagram showing an example of the configuration of a prediction system according to a second embodiment of the present invention. FIG. 6 is a flowchart showing an example of the flow of processing performed by a server according to the second embodiment of the present invention. FIG. 7 is a graph showing the relationship between the number of vacations taken or the number of annual paid leave taken and the level of psychological stress. FIG. 8 is a graph showing the relationship between the average work start time, the average work end time, or the average working hours and the level of psychological stress.

[0013] [First Embodiment] Hereinafter, one embodiment of the present invention will be described in detail.

[0014] (Configuration of prediction system 100) The prediction system 100 in this embodiment is a system that predicts the future level of psychological stress of a subject based on the subject's work style information, measurement information, and behavioral record information. In this embodiment, it is assumed that all workers W to be predicted belong to the same company, and that the prediction system 100 has been introduced to that company. The prediction system 100 predicts the future level of psychological stress of workers, i.e., employees, and can also be called a "worker stress prediction system."

[0015] Fig. 1 is a block diagram showing an example of the configuration of a prediction system 100. As shown in Fig. 1, the prediction system 100 may include a terminal device 10 and a wearable device 20 used by a worker W, a server 30, and an information processing device 60 carried by a person who manages attendance at the company to which the worker W belongs. Although Fig. 1 shows only one worker W, in the prediction system 100, each of multiple workers W carries the terminal device 10 and the wearable device 20.

[0016] The prediction system 100 can predict the future level of psychological stress for each of multiple workers W. Worker W is a worker who can select a desired work style from multiple work styles. The multiple work styles may include telecommuting (working from home). In this case, worker W can select a desired work style from at least telecommuting and in-office work (working by coming into the office). In addition to the above, worker W may also be able to select satellite work (working at a location separate from the office and home) as a work style.

[0017] The terminal device 10 may be a computer such as a smartphone, a tablet terminal, etc. The terminal device 10 includes a control unit 11 that controls each unit of the terminal device 10, a storage unit 12 that stores various data used by the terminal device 10, a communication unit 13 that enables the terminal device 10 to communicate with other devices, an input unit 14 that accepts input operations to the terminal device 10, and a display unit 15 that displays various information.

[0018] The terminal device 10 may be installed with application software (hereinafter referred to as an app) for accepting input of information indicating the actions performed by the worker W and storing the information in the memory unit 12. The terminal device 10 may transmit the information indicating the actions performed by the worker W to the server 30 or the like via the communication unit 13. The terminal device 10 may also accept input from the worker W regarding the actions performed by the worker W and the time when the actions were performed via the input unit 14 via the app. The actions of the worker W may be classified into multiple categories. For example, if the actions of the worker W are classified into 16 categories, the categories may include "sleep," "meals / snacks," "baths," "work / study," and "average viewing time of media such as TV / DVDs." Note that in one aspect of the present invention, at least a portion of the information indicating the behavioral patterns of the worker W may not be input by the worker W. For example, a sensor may detect that worker W is watching media such as TV or DVD, and the sensor may calculate the “average viewing time for media such as TV or DVD” based on the detection result and output it to server 30.

[0019] The terminal device 10 may have a function to acquire information indicating the level of psychological stress of the worker W. For example, the terminal device 10 may have a function to conduct a questionnaire to acquire information indicating the level of psychological stress. In this case, the terminal device 10 may accept input of responses to the questionnaire via the input unit 14. The index indicating the level of psychological stress is not particularly limited, and conventionally known measures such as the K6 score, PHQ-9, HAM-D, and the Simple Occupational Stress Questionnaire (57 items) can be used. In this embodiment, an example will be described in which the K6 score is used as the index indicating the psychological level.

[0020] The wearable device 20 is a device worn on the body of the worker W. The wearable device 20 has a function for measuring data related to the worker W's activity state (amount of activity). Here, the activity state may be the number of steps taken, calories burned, sleep time, conversation time, pulse rate, skin temperature, ultraviolet radiation level, etc. The wearable device 20 may be configured to output information including the measured measurement data to the terminal device 10. The information including the measurement data may include the worker W's activity level and sleep time. The wearable device 20 may be a wearable device worn on the worker W's head, neck, wrist, fingers, chest, abdomen, ankle, etc., for example.

[0021] The server 30 may be a computer. The server 30 includes a control unit 40, a storage unit 50 that stores various data used by the server 30, a communication unit 31 that enables the server 30 to communicate with other devices, and an input unit 32 that accepts input operations for the server 30.

[0022] The control unit 40 controls each unit of the server 30. The control unit 40 includes an information acquisition unit 41, a first extraction unit 42, a first model generation unit 43, and a first prediction unit 44.

[0023] The information acquisition unit 41 acquires information about each worker W from the terminal device 10 or the information processing device 60 via the communication unit 31. Specifically, the information acquisition unit 41 acquires information indicating the work style of each worker W from the information processing device 60 via the communication unit 31. The information includes at least information on the dates on which each worker W worked and information on the dates on which each worker W telecommuted. The information acquisition unit 41 stores the information indicating the work style of each worker W acquired from the information processing device 60 as work style information 51 in the storage unit 50. Furthermore, the information acquisition unit 41 converts each piece of data included in the information indicating the work style of each worker W acquired from the information processing device 60 into data for each predetermined period (one week in this embodiment), and stores information indicating each converted piece of data in the storage unit 50 as work style information 51. For example, the information acquisition unit 41 calculates the telecommuting rate, which is the ratio of telecommuting days to the total number of working days, for each week based on the information on the dates on which each worker W worked and the information on the dates on which they worked from home, which is included in the acquired work style information, and stores the telecommuting rate information indicating the calculated telecommuting rate in the memory unit 50 as one of the work style information 51.

[0024] The information acquisition unit 41 may also acquire information indicating the attendance status of each worker W from the information processing device 60. The attendance status information may include at least one of the following: vacation acquisition date and time, annual paid vacation date and time, work start time, work end time, and working hours for each worker W. The information acquisition unit 41 stores the information indicating the attendance status of each worker W acquired from the information processing device 60 in the storage unit 50 as attendance status information 52. The information acquisition unit 41 also converts each piece of data included in the acquired information indicating the attendance status of each worker W into data for a predetermined period (one week in this embodiment), and stores information indicating the converted pieces of data in the storage unit 50 as attendance status information 52. For example, the information acquisition unit 41 may calculate the number of vacations taken per week based on vacation acquisition date and time information included in the information indicating the attendance status of each worker W acquired from the information processing device 60. The information acquisition unit 41 may then store the calculated number of vacation days taken, which indicates the number of vacation days taken per week, in the storage unit 50 as one piece of attendance status information 52. Similarly, the information acquisition unit 41 may calculate the number of annual paid vacation days taken, the average work start time, the average work end time, the average working hours, the standard deviation of work start times, the standard deviation of work end times, and the standard deviation of working hours for each week. The information acquisition unit 41 may then store each of the calculated data for each week in the storage unit 50 as one piece of attendance status information 52.

[0025] The information acquiring unit 41 also acquires information including data on the activity state of each worker W measured by the wearable device 20 from each of the terminal devices 10 carried by each worker W. The information acquiring unit 41 stores the information including the data on the activity state of each worker W acquired from the terminal device 10 as measurement information 53 in the storage unit 50. The information acquiring unit 41 also converts each piece of data included in the information including the data on the activity state of each worker W acquired from the terminal device 10 into data for a predetermined period (one week in this embodiment), and stores information indicating each piece of converted data in the storage unit 50 as measurement information 53. For example, the information acquiring unit 41 may calculate the average number of steps per day for each week based on step count data information included in the information including the data on the activity state of each worker W acquired from the terminal device 10. The information acquiring unit 41 may then store information indicating the calculated average number of steps per day for each week in the storage unit 50 as one piece of measurement information 53. Furthermore, for example, the information acquisition unit 41 may calculate the average daily sleeping hours for each week based on the sleeping time information included in the information acquired from the terminal device 10. Then, the information acquisition unit 41 may store information indicating the calculated average daily sleeping hours for each week in the storage unit 50 as one piece of measurement information 53.

[0026] The information acquisition unit 41 also acquires information indicating the actions performed by each worker W. For example, the worker W may record the actions performed by the worker W in the terminal device 10 via the app. In this case, the information acquisition unit 41 may acquire information indicating the actions performed by each worker W from the terminal device 10 via the communication unit 31. However, the method by which the information acquisition unit 41 acquires information indicating the actions performed by each worker W is not limited to this. For example, the information acquisition unit 41 may acquire information indicating the actions performed by each worker W from a paper medium on which the worker W recorded his or her own action record. In this case, the action record recorded on the paper medium may be input to the server 30 via the input unit 32 by, for example, an administrator of the prediction system 100. Alternatively, the server 30 may be equipped with a known OCR (Optical Character Recognition) function and may directly read the action record recorded on the paper medium. The information acquisition unit 41 stores the acquired information indicating the actions taken by each worker W in the storage unit 50 as action record information 54 .

[0027] Furthermore, the information acquisition unit 41 may acquire information indicating the attributes of each worker W. The information indicating the attributes includes, for example, gender, occupation, marital status, and age. The information acquisition unit 41 stores the acquired information indicating the attributes in the storage unit 50 as attribute information 55. For example, the information acquisition unit 41 may acquire information indicating the attributes input by the worker W to each terminal device 10 from each terminal device 10 via the communication unit 31. Alternatively, the information indicating the attributes may be input to the server 30 by, for example, an administrator of the prediction system 100. In this case, a predetermined questionnaire inquiring about the attribute information of the worker W may be administered to each worker W in advance, and the administrator may input attribute information 55 indicating the attributes of each worker W to the server 30 based on the responses to the questionnaire.

[0028] The information acquisition unit 41 also acquires information indicating the level of psychological stress of each worker W. The information acquisition unit 41 acquires information indicating the level of psychological stress of each worker W at predetermined intervals (every week in this embodiment) and stores the information indicating the level of psychological stress of each worker W for each week in the storage unit 50 as psychological stress information 56. For example, the information acquisition unit 41 can acquire information indicating the level of psychological stress of each worker W from each terminal device 10 carried by each worker W via the communication unit 31. However, the method by which the information acquisition unit 41 acquires the level of psychological stress of each worker W is not limited thereto. For example, the information acquisition unit 41 may acquire the level of psychological stress of each worker W from a questionnaire to which the worker W answers a questionnaire conducted to acquire the level of psychological stress of each worker W. In this case, for example, the psychological stress information 56 may be created by an administrator of the prediction system 100 based on answers written on paper and input to the server 30 via the input unit 32. Alternatively, the server 30 may be provided with an OCR function and may directly read the answers written on the questionnaire form.

[0029] The first extraction unit 42 extracts a group of similar workers W from among multiple workers W based on the degree of similarity of their working styles to that of a worker W (hereinafter, the worker W will be referred to as the subject O) whose level of psychological stress is to be predicted. The degree of similarity of working styles may be, for example, the difference between the telecommuting rate of the subject O, which is the ratio of telecommuting days to the total number of working days, and the telecommuting rate of the worker W. In this embodiment, a configuration is described in which the first extraction unit 42 extracts a group of similar workers W based on telecommuting rate information indicating the telecommuting rate. In this case, the first extraction unit 42 first reads from the storage unit 50 the telecommuting rate for each of the multiple workers W for a period between the time at which the prediction is made (hereinafter, also referred to as the current time) and a time a first predetermined period before the current time (hereinafter, referred to as the first time). The length of the first period is not particularly limited, and may be, for example, several weeks. In this embodiment, the length of the first period is assumed to be 12 weeks. In this case, the first extraction unit 42 first reads from the storage unit 50 the telecommuting rates of each of the multiple workers W between the current time and a time point 12 weeks prior to the current time.

[0030] Based on the read-out telecommuting rates, the first extraction unit 42 extracts, from the multiple workers W, a group of similar workers whose telecommuting rates during the first period are similar to that of the target person O. For example, the first extraction unit 42 may extract, as the group of similar workers, a predetermined number of workers W in descending order of their telecommuting rates during the first period from that of the target person O. In this case, the first extraction unit 42 may extract the group of similar workers using a k-nearest neighbor algorithm. Alternatively, the first extraction unit 42 may extract, as the group of similar workers, workers W whose telecommuting rates during the first period differ from that of the target person O by a predetermined range.

[0031] As an example, assume that 20 workers W are extracted as a similar worker group. In this case, if the workers W ranked 16th to 25th in terms of similarity to the telecommuting rate of the target person O during the first period have the same telecommuting rate, the first extraction unit 42 first extracts the workers W ranked 1st to 15th in terms of similarity to the telecommuting rate of the target person O during the first period as a similar worker group. Alternatively, the first extraction unit 42 may randomly extract five workers W from the 10 workers W ranked 16th to 25th in terms of similarity to the telecommuting rate of the target person O during the first period, and use these workers as a similar worker group. As a result, a total of 20 similar workers are extracted. In the following description, the workers W included in the similar worker group extracted by the first extraction unit 42 will be referred to as similar workers SW.

[0032] The first model generation unit 43 generates a prediction model (hereinafter referred to as a first prediction model 57) for predicting the degree of psychological stress of the subject O. The first model generation unit 43 generates the first prediction model 57 through machine learning using first information including a first explanatory variable and a first objective variable, which will be described later, as training data. The machine learning model used to create the first prediction model 57 is not particularly limited, and examples thereof include Xgboost and lightGBM. Xgboost stands for eXtreme Gradient Boosting, and is a technique that combines ensemble learning known as gradient boosting with a decision tree. LightGBM is a machine learning framework for gradient boosting based on a decision tree algorithm.

[0033] The first explanatory variables include at least work style information 51, measurement information 53, and behavioral record information 54 for each similar worker SW during the first period. The first explanatory variables may include, as the work style information 51, the telecommuting rate of the similar worker SW during the first period. The first explanatory variables may include, as the measurement information 53, the activity amount and sleep time of the similar worker SW during the first period. The activity amount may include at least one of the number of steps taken, calories burned, sleep time, conversation time, pulse rate, skin temperature, and ultraviolet light level. The first explanatory variables may include, as the behavioral record information 54, the number of times the similar worker SW eats lunch during the first period, information about the similar worker SW's outings, etc. The first explanatory variables may include, as the behavioral record information 54, information about how each similar worker SW spends time outside of work hours during which they can choose to engage in desired activities (hereinafter referred to as leisure time) (hereinafter referred to as leisure time information). The time during which the desired behavior can be selected may be the total time from when the similar worker SW wakes up until when he or she goes to bed, minus the working time. The leisure time information may include, for example, information regarding the length of leisure time.

[0034] The first explanatory variables may further include work style information 51, measurement information 53, and behavior record information 54 for the period between the first time point and a time point (hereinafter referred to as the second time point) that is a second predetermined period before the time point at which the prediction is made (i.e., the current time point) for the subject O (hereinafter referred to as the second period). In this embodiment, the first explanatory variables are described as including the above information for the subject O for the second period. The length of the second period is shorter than the length of the first period, and in this embodiment, the length of the second period is described as 7 weeks. Note that the length of the first period is described as 12 weeks, as described above.

[0035] The first explanatory variables may include attendance status information 52 of each of the similar workers SW in the first period, and attendance status information 52 of the target person O in the second period. The first explanatory variables may also include attribute information 55 of each of the similar workers SW and the target person O.

[0036] The first objective variable includes at least psychological stress information 56 indicating the magnitude of psychological stress in the first period for each of the similar workers SW. In the case where the first explanatory variables include work pattern information 51, measurement information 53, and behavior record information 54 for the subject O, as in this embodiment, the first objective variable includes psychological stress information 56 for the subject O in the second period.

[0037] The first model generation unit 43 reads each piece of information included in the first information (i.e., the first explanatory variable and the first objective variable) from the storage unit 50, and generates a first prediction model 57 by machine learning using the first information constituted by the read pieces of information as learning data. The first model generation unit 43 stores the generated first prediction model 57 in the storage unit 50.

[0038] The first prediction unit 44 predicts the level of psychological stress of the subject O from the first period onward by inputting input data including at least the work pattern information 51, measurement information 53, and behavioral record information 54 of the subject O into the first prediction model 57 generated by the first model generation unit 43. In the present embodiment, the first prediction unit 44 predicts the level of psychological stress of the subject O from the first period onward by inputting data including the above-mentioned various information for the period between the second time point and the time point at which the prediction is made (i.e., the current time) (hereinafter, this period will be referred to as the third period) as input data into the first prediction model 57. In the present embodiment, as described above, the length of the first period is 12 weeks and the length of the second period is 7 weeks, so the length of the third period is 5 weeks. The first prediction unit 44 may predict the level of psychological stress of the subject O, for example, several days to several weeks after the current time point (i.e., the latest time point in the first period). In this embodiment, the first prediction unit 44 predicts the magnitude of psychological stress of the subject O one week from the present time.

[0039] The first prediction unit 44 may use the K6 score as an index indicating the magnitude of psychological stress. In this case, the value of the K6 score may be set as the objective variable. As another example, the objective variable may be set to classify into multiple classes based on the value of the K6 score. For example, the objective variable may be set to classify into multiple classes, such as class 0 for a K6 score less than 5, class 1 for a K6 score between 5 and 9, class 2 for a K6 score between 9 and 13, and class 3 for a K6 score of 13 or greater. As another example, the objective variable may be set to classify into two classes, such as class 0 for a K6 score less than 5 and class 1 for a K6 score of 5 or greater. Instead of the K6 score, other indicators (e.g., PHQ-9, HAM-D, etc.) may be used as an indicator indicating the magnitude of psychological stress. In this embodiment, the K6 score is described as being used as an indicator indicating the magnitude of psychological stress.

[0040] (Processing Performed by Server 30) The flow of processing performed by the server 30 will be described below with reference to Fig. 2. Fig. 2 is a flowchart showing an example of the flow of processing performed by the server 30.

[0041] First, the information acquisition unit 41 acquires information about each worker W (step S1). Specifically, the information acquisition unit 41 acquires, from the terminal device 10 or the information processing device 60, information indicating the work style of each worker W, information indicating the attendance status of each worker W, information including data on the activity status of each worker W measured by the wearable device 20, information indicating the actions taken by each worker W, information indicating the attributes of each worker W, and information indicating the level of psychological stress of each worker W.

[0042] The information acquisition unit 41 stores each piece of acquired information in the storage unit 50 (step S2). The information acquisition unit 41 also calculates the telecommuting rate for each worker W on a weekly basis based on the information on the dates on which each worker W worked and the information on the dates on which each worker W worked from home, which are included in the acquired work style information, and stores telecommuting rate information indicating the calculated telecommuting rate in the storage unit 50 as one piece of work style information 51 (step S3).

[0043] Next, the first extraction unit 42 extracts a group of similar workers from the multiple workers W based on the degree of similarity of their working styles to that of the target person O (step S4, first extraction step). Specifically, the first extraction unit 42 reads out the telecommuting rate for each of the multiple workers W during the first period from the storage unit 50. Based on the read telecommuting rates, the first extraction unit 42 extracts a group of similar workers from the multiple workers W whose telecommuting rates are similar to that of the target person O during the first period.

[0044] Next, the first model generation unit 43 generates a first prediction model 57 for predicting the level of psychological stress of the subject O (step S5, first model generation step). Specifically, the first model generation unit 43 generates the first prediction model 57 by machine learning using, as learning data, first information including at least work style information 51, measurement information 53 of each worker W, behavior record information 54 of each worker W, and psychological stress information 56 of each worker W, for a first period, for each worker W included in the group of similar workers extracted by the first extraction unit 42. The first model generation unit 43 reads each piece of information to be used as learning data from the storage unit 50.

[0045] Next, the first prediction unit 44 predicts the magnitude of psychological stress of the subject O from the first period onwards by inputting input data into the first prediction model 57 generated by the first model generation unit 43 (step S6, first prediction step). Specifically, the first prediction unit 44 predicts the magnitude of psychological stress of the subject O one week from the present time by inputting input data including at least the work style information 51, measurement information 53, and behavior record information 54 of the subject O for the third period into the first prediction model 57.

[0046] (Effects of prediction system 100) The inventors discovered that by using a first prediction model 57 generated by machine learning using as learning data first information that includes at least information on each similar worker SW included in a group of similar workers whose working styles are highly similar to that of the subject O, it is possible to accurately predict the level of psychological stress of the subject O from the first period onwards, and completed the prediction system 100 of this embodiment.

[0047] Here, we will explain the results of a prediction accuracy test using a prediction model generated by a conventional method and using the first prediction model 57 generated by the method described in this embodiment. In this prediction accuracy test, 192 employees W working at the same company were tested as subjects.

[0048] When the conventional prediction model was used, a prediction model was first generated by machine learning using, as training data, information including work style information 51, measurement information 53, behavior record information 54, and psychological stress information 56 of 192 workers W for the period between 12 weeks before the time of prediction and 5 weeks before the time of prediction. Next, information including work style information 51, measurement information 53, and behavior record information 54 of each worker W for the period between 5 weeks before the time of prediction and the time of prediction was input as input data to the generated prediction model, thereby predicting the K6 score for each worker W one week after the time of prediction.

[0049] Next, a case will be described where the first prediction model 57 generated by the method described in this embodiment is used. In this case, a first prediction model 57 is generated for each of the 192 workers W. Here, a method will be described in which the first prediction model 57 is generated using one worker W out of the 192 workers W as the subject O. First, 20 similar workers SW were extracted as a similar worker group from the 191 workers W excluding the subject O in order of the degree of similarity between the telecommuting rate of the subject O and the telecommuting rate of the period between 12 weeks before the time of prediction and the time of prediction (i.e., the first period). Next, a first prediction model 57 for the subject O was generated by machine learning using as training data first information including (1) the work style information 51, measurement information 53, behavior record information 54, and psychological stress information 56 for the extracted 20 similar workers SW during the first period, and (2) the work style information 51, measurement information 53, behavior record information 54, and psychological stress information 56 for the subject O during the period between 12 weeks before the time of the prediction and 5 weeks before the time of the prediction (i.e., the second period). First prediction models 57 were also generated for the other 191 workers W using a similar method. Next, information including work style information 51, measurement information 53, and behavioral record information 54 for the corresponding worker W during the period between the time five weeks before the time the prediction was made and the time the prediction was made (i.e., the third period) was input as input data into each of the first prediction models 57 generated for each of the 192 workers W, thereby predicting the K6 score for each worker W one week after the time the prediction was made.

[0050] 3 shows the prediction results obtained using a conventional prediction model and the first prediction model 57 generated by the method described in this embodiment, along with ROC (Receiver Operating Characteristic) curves generated based on the actual K6 scores (i.e., the true values ​​of the K6 scores) of each worker W one week after the prediction. As shown in FIG. 3 , the ROC curve G1 obtained using the first prediction model 57 generated by the method described in this embodiment has a higher TPR (True Positive Rate) at most FPRs (False Positive Rates) and higher prediction accuracy than the ROC curve G2 obtained using the conventional prediction model. When the AUC (Area Under the ROC Curve) was calculated using each ROC curve shown in FIG. 3 , the AUC was 0.760 when the conventional prediction model was used, while the AUC was 0.852 when the first prediction model 57 generated by the method described in this embodiment was used. The criteria for evaluating AUC are described in "JASwets "Measuring the accuracy of diagnostic systems" Science. 1988 Jun 3;240(4857):1285-93.doi: 10.1126 / science.3287615." as follows: 0.50 to 0.60 is a "prediction at chance," 0.60 to 0.70 is a "poor prediction," 0.70 to 0.80 is a "fair prediction," 0.80 to 0.90 is a "good prediction," and 0.90 to 1.00 is an "excellent prediction." It was shown that when the first prediction model 57 generated by the method described in this embodiment was used, the prediction accuracy was high.

[0051] The reason why prediction accuracy is high when using the first prediction model 57 generated using the method described in this embodiment is thought to be because the first prediction model 57 is generated using training data that includes information about each similar worker SW included in the group of similar workers whose working styles are highly similar to those of the subject O.

[0052] Here, because the level of psychological stress is similar among workers with highly similar working styles, it is conceivable that the accuracy of the prediction of the level of psychological stress will be high if a prediction model generated using information on workers with highly similar working styles is used. However, it should be noted that this idea is inappropriate. This will be explained using Figure 4. Figure 4 is a graph showing the relationship between the rate of telecommuting and the level of psychological stress. The graph shown in Figure 4 has the rate of telecommuting over a specified period on the horizontal axis and the level of psychological stress on the vertical axis. The graph shown in Figure 4 was created using data from 192 workers W who were subjects in the above-mentioned prediction accuracy test. Calculations made from the graph shown in Figure 4 revealed that the correlation coefficient between the rate of telecommuting and the level of psychological stress was -0.032, indicating a small correlation between the rate of telecommuting and the level of psychological stress.

[0053] In the prediction system 100 of this embodiment, the server 30 may output the prediction result of the level of psychological stress of the subject O predicted by the first prediction unit 44 to the terminal device 10 carried by the subject O via the communication unit 31. The terminal device 10 may receive the prediction result of the level of psychological stress of the subject O from the server 30 via the communication unit 13 and display a display screen on the display unit 15 to notify the subject O of the prediction result of the level of psychological stress. This allows the subject O to recognize the level of his / her psychological stress from the first period onwards. Furthermore, the terminal device 10 may display a display screen on the display unit 15 to display an alert to the subject O when the predicted value of the level of psychological stress is greater than a predetermined value. This allows the subject O to recognize that his / her psychological stress will increase from the first period onwards.

[0054] In the prediction system 100 of this embodiment, the server 30 may output the prediction result of the level of psychological stress of the subject O predicted by the first prediction unit 44 to an information processing device held by a manager (e.g., the superior of worker W) who manages worker W in the organization (e.g., department, group, etc.) to which worker W belongs, via the communication unit 31. This allows the manager to know the number of workers W in the organization who are predicted to have a high level of psychological stress, thereby enabling the manager to manage the organization appropriately.

[0055] In the prediction system 100 of this embodiment, the server 30 may output the prediction result of the level of psychological stress of the subject O predicted by the first prediction unit 44 to an information processing device owned by a human resources manager who manages the human resources of the company to which the worker W belongs, via the communication unit 31. This allows the human resources manager to appropriately decide on company measures and evaluate the workers W based on the number of workers W predicted to have high psychological stress in the company.

[0056] In the prediction system 100 of this embodiment, the server 30 may output the prediction result of the level of psychological stress of the subject O predicted by the first prediction unit 44 to an information processing device owned by an industrial physician of the company to which the worker W belongs, via the communication unit 31. This allows the industrial physician to take appropriate measures, such as interviewing the worker W who is predicted to have high psychological stress.

[0057] In the prediction system 100 of this embodiment, when extracting a group of similar workers, the first extraction unit 42 extracts multiple similar workers SW whose telecommuting rates during the first period are similar to those of the subject O during the first period, and generates a first prediction model 57 using first information including work style information 51, measurement information 53, behavior record information 54, and psychological stress information 56 for each similar worker SW during the first period. That is, the period during which input data to be input into the first prediction model 57 was acquired is the same as the period during which data included in the first information used to generate the first prediction model 57 was acquired. This reduces the influence of differences in psychological stress due to the environment (e.g., season, age, etc.) at the time the data was acquired. As a result, the prediction accuracy of the level of psychological stress can be improved. Note that in one aspect of the prediction system 100 of this embodiment, when extracting a group of similar workers, the first extraction unit 42 may extract similar workers SW whose telecommuting rates during any period other than the first period are similar to those of the subject O during the first period. The first model generation unit 43 may then generate the first prediction model 57 using first information including the work pattern information 51, measurement information 53, behavior record information 54, and psychological stress information 56 of the similar worker SW for the arbitrary period. In other words, the period during which input data to be input into the first prediction model 57 was obtained may be different from the period during which data included in the first information used to generate the first prediction model 57 was obtained. The length of the arbitrary period may be the same as the first period, or may be a period of a different length from the first period.

[0058] In this embodiment, all of the workers W belong to the same company, but the present invention is not limited to this. The prediction system 100 of one aspect of the present invention may be applied to multiple workers W who belong to different companies, or may be applied to a group of workers in which some workers W work for different companies, as long as the workers W can select their desired work style from multiple work styles.

[0059] [Embodiment 2] Another embodiment of the present invention will be described below. For ease of explanation, the same reference numerals will be used to designate components having the same functions as those described in the above embodiment, and the description thereof will not be repeated.

[0060] The prediction system 100 in the first embodiment predicted the level of psychological stress of the subject O using a first prediction model 57 generated by machine learning using as learning data first information including various information of a group of similar workers extracted based on the degree of similarity of their working styles to that of the subject O. In contrast, the prediction system 100A in the present embodiment predicts the level of psychological stress of the subject O using a prediction model generated by machine learning using as learning data second information including various information of a group of similar workers extracted based on the degree of similarity of their working styles to that of the subject O.

[0061] Fig. 5 is a block diagram showing an example of the configuration of a prediction system 100A. As shown in Fig. 5, the prediction system 100A includes a server 30A instead of the server 30 in the first embodiment. The server 30A includes a control unit 40A and a storage unit 50A instead of the control unit 40 and the storage unit 50 in the first embodiment. The control unit 40A includes a second extraction unit 72, a second model generation unit 73, and a second prediction unit 74 instead of the first extraction unit 42, the first model generation unit 43, and the first prediction unit 44 in the first embodiment.

[0062] The second extraction unit 72 extracts a group of similar workers from multiple workers W based on the degree of similarity of their attendance status to that of a subject O, whose level of psychological stress is to be predicted. The attendance status may be at least one of the number of vacation days taken, the number of annual paid vacation days taken, the average work start time, the average work end time, and the average working hours. In this case, the degree of similarity of the attendance status may be, for example, the difference between the number of vacation days taken by subject O in a first period and the number of vacation days taken by worker W in a predetermined period (however, the same length as the first period). In this embodiment, a configuration is described in which the second extraction unit 72 extracts a group of similar workers based on vacation count information indicating the number of vacation days taken. In this case, the second extraction unit 72 first reads from the storage unit 50A the number of vacation days taken for each worker W during the period between the time at which the prediction is made (i.e., the current time) and a time three predetermined periods before the current time (hereinafter referred to as the third time) (in this embodiment, this period is referred to as the first period).

[0063] Based on the read-out number of vacation days taken, the second extraction unit 72 extracts, from the multiple workers W, a group of similar workers whose number of vacation days taken during the first period is similar to that of the target worker O. For example, the second extraction unit 72 may extract, as the group of similar workers, a predetermined number of workers W in descending order of the number of vacation days taken during the first period that is closest to that of the target worker O. In this case, the second extraction unit 72 may extract the group of similar workers using a k-nearest neighbor algorithm. Alternatively, the second extraction unit 72 may extract, as the group of similar workers, workers W whose number of vacation days taken during the first period differs from the number of vacation days taken during the first period by the target worker O by a predetermined range. In the following description, the workers W included in the group of similar workers extracted by the second extraction unit 72 will be referred to as similar workers SW.

[0064] The second model generation unit 73 generates a prediction model (hereinafter referred to as the second prediction model 77) for predicting the magnitude of psychological stress of the subject O. The second model generation unit 73 generates the second prediction model 77 by machine learning using second information including a second explanatory variable and a second objective variable, which will be described later, as training data. The machine learning model used to create the second prediction model 77 is not particularly limited, and may be, for example, Xgboost or lightGBM.

[0065] The second explanatory variables include at least the attendance status information 52, measurement information 53, and behavioral record information 54 for each similar worker SW during the first period. The second explanatory variables may include, as the attendance status information 52, information indicating at least one of the number of vacations taken, the number of annual paid vacations taken, the average start time of work, the average end time of work, and the average working hours of each similar worker SW during the first period. The second explanatory variables may include, as the measurement information 53, the activity amount and sleep time of each similar worker SW during the first period. The activity amount may include at least one of the number of steps taken, calories burned, sleep time, conversation time, pulse rate, skin temperature, and ultraviolet light level. The second explanatory variables may include, as the behavioral record information 54, the number of times each similar worker SW eats lunch during the first period, information about the similar worker SW's outings, and the like. The second explanatory variables may also include, as the behavioral record information, leisure time information for each similar worker SW. The leisure time information may include, for example, information regarding the length of leisure time.

[0066] The second explanatory variables may include the attendance status information 52, the measurement information 53, and the behavior record information 54 for the period between the third time point and a time point (hereinafter referred to as the fourth time point) four predetermined periods before the time point at which the prediction is made (i.e., the present time) (in this embodiment, this period is referred to as the second period). In this embodiment, the second explanatory variables are described as including the above information for the subject O for the second period. The length of the second period is shorter than the length of the first period, and in this embodiment, the length of the second period is described as 7 weeks. Note that the length of the first period is described as 12 weeks, as described above.

[0067] The second explanatory variables may include work style information 51 of each similar worker SW in the first period, and work style information 51 of the target person O in the second period. The second explanatory variables may also include attribute information 55 of each similar worker SW and the target person O.

[0068] The second objective variable includes at least psychological stress information 56 indicating the magnitude of psychological stress during the first period for each of the similar workers SW. In the case where the second explanatory variables include attendance status information 52, measurement information 53, and behavior record information 54 for the subject O, as in this embodiment, the second objective variable includes psychological stress information 56 for the subject O during the second period.

[0069] The second model generation unit 73 reads each piece of information included in the second information (i.e., the second explanatory variable and the second objective variable) from the storage unit 50A, and generates a second prediction model 77 by machine learning using the second information constituted by the read pieces of information as learning data. The second model generation unit 73 stores the generated second prediction model 77 in the storage unit 50A.

[0070] The second prediction unit 74 predicts the level of psychological stress of the subject O from the first period onward by inputting input data including at least the attendance status information 52, measurement information 53, and behavior record information 54 of the subject O into the second prediction model 77 generated by the second model generation unit 73. In the present embodiment, the second prediction unit 74 predicts the level of psychological stress of the subject O from the first period onward by inputting data including the above-mentioned various information for the period between the second time point and the time point at which the prediction is made (i.e., the current time) (in the present embodiment, this period is referred to as the third period) as input data into the second prediction model 77. In the present embodiment, as described above, the length of the first period is 12 weeks and the length of the second period is 7 weeks, so the length of the third period is 5 weeks. The second prediction unit 74 may predict the level of psychological stress of the subject O, for example, several days to several weeks after the current time point (i.e., the latest time point in the first period). In this embodiment, the second prediction unit 74 predicts the level of psychological stress of the subject O one week from the present time.

[0071] (Processing Performed by Server 30A) The flow of processing performed by the server 30A will be described below with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the flow of processing performed by the server 30A.

[0072] First, in the process performed by the server 30A, steps S1 and S2 described in the first embodiment are performed.

[0073] Next, the second extraction unit 72 extracts a group of similar workers from the multiple workers W based on the degree of similarity of their attendance situations to that of the target person O (step S13, second extraction step). Specifically, the second extraction unit 72 reads out the number of vacation days taken for each of the multiple workers W during the first period from the storage unit 50A. Based on the read number of vacation days taken, the second extraction unit 72 extracts from the multiple workers W a group of similar workers whose number of vacation days taken during the first period is similar to that of the target person O.

[0074] Next, the second model generation unit 73 generates a second prediction model 77 for predicting the level of psychological stress of the subject O (step S14, second model generation step). Specifically, the second model generation unit 73 generates the second prediction model 77 by machine learning using, as training data, second information including at least the attendance status information 52, measurement information 53 of each worker W, behavior record information 54 of each worker W, and psychological stress information 56 of each worker W, for the first period, of each worker W included in the group of similar workers extracted by the second extraction unit 72. The second model generation unit 73 reads each piece of information to be used as training data from the storage unit 50A.

[0075] Next, the second prediction unit 74 predicts the magnitude of psychological stress of the subject O from the first period onwards by inputting input data into the second prediction model 77 generated by the second model generation unit 73 (step S15, second prediction step). Specifically, the second prediction unit 74 predicts the magnitude of psychological stress of the subject O one week from the present time by inputting input data including at least the attendance status information 52, measurement information 53, and behavior record information 54 of the subject O for the third period into the second prediction model 77.

[0076] (Effects of prediction system 100A) The inventors discovered that by using a second prediction model 77 generated by machine learning using as learning data second information that includes at least information on each similar worker SW included in a group of similar workers whose attendance status is highly similar to that of the subject O, it is possible to accurately predict the level of psychological stress of the subject O from the first period onwards, and completed the prediction system 100A in this embodiment.

[0077] Here, we will explain the results of a prediction accuracy test using a prediction model generated by a conventional method and using the second prediction model 77 generated by the method described in this embodiment. In this prediction accuracy test, 192 employees W working at the same company were tested as subjects.

[0078] The conventional prediction model is the conventional prediction model described in embodiment 1. As described in embodiment 1, when the conventional prediction model was used, the AUC was 0.760.

[0079] When the second prediction model 77 generated using the method described in this embodiment was used, the second prediction model 77 was generated using the same method as the first prediction model 57 in the prediction accuracy test described in embodiment 1, except that instead of using the teleworking rate, the number of vacation days taken, the number of annual paid vacation days taken, the average start time of work, the average end time of work, or the average working hours were used when extracting a group of similar workers.

[0080] When the second prediction model 77 generated by the method described in this embodiment was used, the AUC was calculated in the same manner as in the prediction accuracy test described in embodiment 1. As a result, when extracting a group of similar workers, the AUC was 0.852 when the number of vacation days taken was used, the AUC was 0.838 when the number of annual paid vacation days taken was used, the AUC was 0.829 when the average work start time was used, the AUC was 0.832 when the average work end time was used, and the AUC was 0.830 when the average working hours were used, indicating that the second prediction model 77 generated by the method described in this embodiment had high prediction accuracy.

[0081] The reason why prediction accuracy is higher when using the second prediction model 77 generated using the method described in this embodiment is thought to be because the second prediction model 77 is generated using training data containing information about each similar worker SW included in the group of similar workers whose attendance status is highly similar to that of the subject O.

[0082] Here, it is conceivable that workers with similar attendance records have similar levels of psychological stress, leading to higher accuracy in predicting the level of psychological stress. However, it should be noted that this idea is inappropriate. This will be explained using Figures 7 and 8. Figure 7 is a graph showing the relationship between the number of vacations or the number of annual paid vacations taken and the level of psychological stress. The graph designated by reference numeral 701 in Figure 7 has the number of vacations taken in a specified period on the horizontal axis and the K6 score on the vertical axis. The graph designated by reference numeral 702 in Figure 7 has the number of annual paid vacations taken in a specified period on the horizontal axis and the K6 score on the vertical axis. Figure 8 is a graph showing the relationship between the average work start time, average work end time, or average working hours and the level of psychological stress. The graph designated by reference numeral 801 in Figure 8 has the average work start time in a specified period on the horizontal axis and the K6 score on the vertical axis. In the graph designated by reference numeral 801, the average work start time is shown as the time elapsed from midnight. The unit of the values ​​on the horizontal axis of the graph indicated by reference numeral 801 is minutes. The graph indicated by reference numeral 802 in FIG. 8 is a graph in which the horizontal axis represents the average work end time in a predetermined period and the vertical axis represents the K6 score. In the graph indicated by reference numeral 802, the average work end time is shown as the time elapsed from midnight. The unit of the values ​​on the horizontal axis of the graph indicated by reference numeral 802 is minutes. The graph indicated by reference numeral 803 in FIG. 8 is a graph in which the horizontal axis represents the average work hours in a predetermined period and the vertical axis represents the K6 score. The unit of the values ​​on the horizontal axis of the graph indicated by reference numeral 803 is minutes. As shown in FIGS. 7 and 8, it can be seen that there is little correlation with the attendance situation K6 score. Specifically, the coefficient of determination R of the regression equation of the graph indicated by reference numeral 701 in FIG. 7 is ２ is 2.82 x 10 －４ The correlation coefficient between the number of vacations taken and the K6 score was 0.017. The coefficient of determination R ２ is 4.31 x 10 －４ The correlation coefficient between the number of annual paid holidays taken and the K6 score was 0.021. The coefficient of determination R ２ is 4.08 x 10 －３The correlation coefficient between the average work start time and the K6 score was 0.063. The coefficient of determination R ２ is 1.45 x 10 －３ The correlation coefficient between the average work end time and the K6 score was 0.038. The coefficient of determination R ２ is 5.50 x 10 －６ The correlation coefficient between average working hours and K6 score was −0.002.

[0083] In the prediction system 100A of this embodiment, when extracting a group of similar workers, the second extraction unit 72 extracts multiple similar workers SW whose number of vacation days taken during the first period is similar to that of the subject O. The second extraction unit 72 then generates a second prediction model 77 using second information for each similar worker SW, including attendance status information 52, measurement information 53, behavior record information 54, and psychological stress information 56 for the first period. That is, the period during which input data to the second prediction model 77 was acquired is the same as the period during which data included in the second information used to generate the second prediction model 77 was acquired. This reduces the influence of differences in psychological stress due to the environment (e.g., season, age, etc.) at the time of data acquisition. As a result, the prediction accuracy of the level of psychological stress can be improved. Note that in one aspect of the prediction system 100A of this embodiment, when extracting a group of similar workers, the second extraction unit 72 may extract similar workers SW whose number of vacation days taken during any period other than the first period is similar to that of the subject O during the first period. The second model generation unit 73 may then generate the second prediction model 77 using second information including the attendance status information 52, measurement information 53, behavior record information 54, and psychological stress information 56 of the similar worker SW for the arbitrary period. In other words, the period during which input data to be input into the second prediction model 77 was obtained may be different from the period during which data included in the second information used to generate the second prediction model 77 was obtained. The length of the arbitrary period may be the same as or different from the second period.

[0084] [Example of implementation using software] The functions of server 30 or server 30A (hereinafter referred to as "device") can be realized by a program that causes a computer to function as the device, and a program that causes a computer to function as each control block of the device (particularly each part included in control unit 40 or control unit 40A).

[0085] In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using the control device and storage device.

[0086] The program may be non-transitory and may be recorded on one or more computer-readable recording media. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

[0087] Furthermore, some or all of the functions of the control blocks can be realized by logic circuits. For example, an integrated circuit in which a logic circuit that functions as each of the control blocks is formed is also included in the scope of the present invention. In addition, the functions of the control blocks can also be realized by, for example, a quantum computer.

[0088] The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

[0089] (Summary) A prediction system according to aspect 1 of the present disclosure comprises: a first extraction unit that extracts a group of similar workers from a plurality of workers based on the degree of similarity of their work styles with the work style of a subject person during a first period; a first model generation unit that generates a first prediction model by machine learning using, as learning data, first information that includes at least: work style information indicating the work style of each worker included in the group of similar workers; measurement information including the activity level and sleep duration of each worker; behavior record information that records the time of behavior performed by each worker by behavior type; and psychological stress information that indicates the level of psychological stress of each worker; and a first prediction unit that predicts the level of psychological stress of the subject person from the first period onwards by inputting input data that includes at least the work style information, the measurement information, and the behavior record information of the subject person into the first prediction model.

[0090] A prediction system according to aspect 2 of the present disclosure may be configured as follows: in aspect 1 above, the plurality of work styles may include at least telecommuting, and the first extraction unit may extract the group of similar workers based on telecommuting rate information indicating the proportion of telecommuting days out of the total number of working days.

[0091] In the prediction system according to aspect 3 of the present disclosure, in the above-mentioned aspect 1 or 2, the first model generation unit may use the first information including the work style information, the measurement information, the behavior record information, and the psychological stress information of each worker during the first period as the learning data.

[0092] A prediction system according to a fourth aspect of the present disclosure includes a second extraction unit that extracts a group of similar workers from a plurality of workers based on the degree of similarity of their attendance status with the attendance status of a subject person during a first period; a second model generation unit that generates a second prediction model by machine learning using second information as learning data, the second information including at least attendance status information indicating the attendance status of each worker included in the group of similar workers, measurement information including the activity level and sleep time of each worker, behavior record information that records the time of behavior performed by each worker by behavior type, and psychological stress information indicating the level of psychological stress of each worker; and a second prediction unit that predicts the level of psychological stress of the subject person from the first period onwards by inputting input data including at least the attendance status information, the measurement information, and the behavior record information of the subject person into the second prediction model.

[0093] In the prediction system according to aspect 5 of the present disclosure, in aspect 4 above, the attendance status information may be information indicating at least one of the number of vacations taken, the number of annual paid vacations taken, the average start time of work, the average end time of work, and the average working hours.

[0094] In the prediction system according to aspect 6 of the present disclosure, in the above-mentioned aspect 4 or 5, the second model generation unit may use the second information including the attendance status information, the measurement information, the behavior record information, and the psychological stress information of each worker during the first period as the learning data.

[0095] In a prediction system according to aspect 7 of the present disclosure, in any of aspects 1 to 6 above, the behavioral record information may include information regarding how each worker spends time that is not working hours and during which desired behaviors can be selected.

[0096] A prediction method according to aspect 8 of the present disclosure includes: a first extraction step of extracting a group of similar workers from a plurality of workers based on the degree of similarity of their work styles with the work style of a subject person during a first period; a first model generation step of generating a first prediction model by machine learning using, as learning data, first information including at least: work style information indicating the work style of each worker included in the group of similar workers; measurement information including the activity level and sleep duration of each of the workers; behavior record information that records the time of behavior performed by each of the workers by behavior type; and psychological stress information indicating the level of psychological stress of each of the workers; and a first prediction step of predicting the level of psychological stress of the subject person from the first period onwards by inputting input data including at least the work style information, the measurement information, and the behavior record information of the subject person into the first prediction model.

[0097] A prediction method according to a ninth aspect of the present disclosure includes: a second extraction step of extracting a group of similar workers from a plurality of workers based on the degree of similarity of their attendance status with the attendance status of a subject person during a first period; a second model generation step of generating a second prediction model by machine learning using, as learning data, second information including at least: attendance status information indicating the attendance status of each worker included in the group of similar workers; measurement information including the activity level and sleep time of each worker; behavior record information recording the time of behavior performed by each worker by behavior type; and psychological stress information indicating the level of psychological stress of each worker; and a second prediction step of predicting the level of psychological stress of the subject person from the first period onwards by inputting input data including at least the attendance status information, the measurement information, and the behavior record information of the subject person into the second prediction model.

[0098] A program according to aspect 10 of the present disclosure is a program for causing a computer to function as a prediction system according to any one of aspects 1 to 3 above, and is a program for causing a computer to function as the first extraction unit, the first model generation unit, and the first prediction unit.

[0099] A program according to aspect 11 of the present disclosure is a program for causing a computer to function as a prediction system according to any one of aspects 4 to 6, and is a program for causing a computer to function as the second extraction unit, the second model generation unit, and the second prediction unit.

[0100] 42 First extraction unit 43 First model generation unit 44 First prediction unit 51 Work style information 52 Attendance status information 53 Measurement information 54 Behavioral record information 56 Psychological stress information 57 First prediction model 72 Second extraction unit 73 Second model generation unit 74 Second prediction unit 77 Second prediction model 100, 100A Prediction system

Claims

1. a first extraction unit that extracts a group of similar workers for each target person from a plurality of workers based on the degree of similarity of their working style to the working style of the target person during a first period; a first model generation unit that generates, for each subject, a first prediction model for predicting the level of psychological stress of the subject by machine learning using first information as learning data, the first information including at least: work style information indicating the work style of each worker included in the group of similar workers; measurement information including the activity amount and sleep time of each worker; behavior record information that records the time of behavior performed by each worker for each behavior type; and psychological stress information indicating the level of psychological stress of each worker; a first prediction unit that predicts the magnitude of psychological stress of the subject from the first period onward for each subject by inputting input data including at least the work style information, the measurement information, and the behavior record information of the subject into the first prediction model; A prediction system comprising:

2. The plurality of work styles includes at least telecommuting, the first extraction unit extracts the group of similar workers based on teleworking rate information indicating the ratio of teleworking days to the total number of working days; The prediction system of claim 1 .

3. the first model generation unit uses the first information, including the work style information, the measurement information, the behavior record information, and the psychological stress information of each worker during the first period, as the learning data; The prediction system of claim 1 .

4. a second extraction unit that extracts a group of similar workers for each subject from a plurality of workers based on the similarity of their attendance status to the attendance status of the subject during a first period; a second model generation unit that generates, for each subject, a second prediction model for predicting the level of psychological stress of the subject by machine learning using second information as learning data, the second information including at least attendance status information indicating the attendance status of each worker included in the group of similar workers, measurement information including the activity amount and sleep time of each worker, behavior record information that records the time of behavior performed by each worker for each behavior type, and psychological stress information indicating the level of psychological stress of each worker; a second prediction unit that predicts the magnitude of psychological stress of the subject from the first period onward for each subject by inputting input data including at least the attendance status information, the measurement information, and the behavior record information of the subject into the second prediction model; A prediction system comprising:

5. The attendance status information is information indicating at least one of the number of vacations taken, the number of annual paid vacations taken, the average start time of work, the average end time of work, and the average working hours. The prediction system of claim 4 .

6. the second model generation unit uses the second information, including the attendance status information, the measurement information, the behavior record information, and the psychological stress information of each of the workers in the first period, as the learning data; The prediction system of claim 4 .

7. The activity record information includes information about how each worker spends time that is not working hours and during which a desired activity can be selected. The prediction system according to claim 1 or 4.

8. a first extraction step of extracting a group of similar workers for each subject from a plurality of workers based on the degree of similarity of their working style to the working style of the subject during a first period; a first model generation step of generating, for each subject, a first prediction model for predicting the level of psychological stress of the subject by machine learning using, as learning data, first information including at least: work style information indicating the work style of each worker included in the group of similar workers; measurement information including the activity amount and sleep time of each worker; behavior record information recording the time of behavior performed by each worker for each behavior type; and psychological stress information indicating the level of psychological stress of each worker; a first prediction step of predicting the magnitude of psychological stress of the subject from the first period onward for each subject by inputting input data including at least the work style information, the measurement information, and the behavior record information of the subject into the first prediction model; A forecasting method including:

9. a second extraction step of extracting a group of similar workers for each subject from a plurality of workers based on the degree of similarity of their attendance status with the attendance status of the subject during the first period; a second model generation step of generating, for each subject, a second prediction model for predicting the level of psychological stress of the subject by machine learning using second information as learning data, the second information including at least attendance status information indicating the attendance status of each worker included in the group of similar workers, measurement information including the activity amount and sleep time of each worker, behavior record information recording the time of behavior performed by each worker for each behavior type, and psychological stress information indicating the level of psychological stress of each worker; a second prediction step of predicting the magnitude of psychological stress of the subject from the first period onward for each subject by inputting input data including at least the attendance status information, the measurement information, and the behavior record information of the subject into the second prediction model; A forecasting method including:

10. A program for causing a computer to function as the prediction system according to claim 1, the program causing a computer to function as the first extraction unit, the first model generation unit, and the first prediction unit.

11. A program for causing a computer to function as the prediction system according to claim 4, the program causing a computer to function as the second extraction unit, the second model generation unit, and the second prediction unit.