Prediction device, prediction method, and prediction program

Abstract

To provide a prediction device capable of further improving accuracy in a classification task, a prediction method, and a prediction program.SOLUTION: A prediction device comprises an acquisition unit and a learning unit. The acquisition unit acquires a first dataset composed of a combination of a first regression value output when arbitrary data is input to a first regression model and a first correction value obtained by tuning the first regression value. The learning unit generates an improved model, which is another regression model for classifying data of the same type as the first regression model by learning the first correction values in the first dataset as ground truth data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a prediction device, a prediction method, and a prediction program.

Background Art

[0002] Predictive or classification processing using a learning model generated by machine learning or the like has been actively used.

[0003] As a technique related to machine learning, for example, a technique has been proposed in which a plurality of regression models are created and the model with the highest accuracy is selected and displayed (for example, Patent Document 1). Further, as another technique related to machine learning, a technique for efficiently extracting teacher data has been proposed (for example, Patent Document 2).

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] The technique according to Patent Document 1 above is considered an effective means for data analysis and the like because it creates different models for each feature amount. However, in the use of machine learning, when the object or purpose of predictive or classification processing is already determined, the step of creating a plurality of different models is often unnecessary.

[0006] Further, the technique according to Patent Document 2 above automates the labeling of teacher data and improves the accuracy of the model by replicating the teacher data, but there is a limit to improving the accuracy only by replicating the teacher data.

[0007] Therefore, the present disclosure proposes a prediction device, a prediction method, and a prediction program capable of achieving further improvement in accuracy in a classification task.

Means for Solving the Problems

[0008] In order to solve the above problems, a prediction device according to the present disclosure includes an acquisition unit that acquires a first data set including a combination of a first regression value output when arbitrary data is input to a first regression model and a first correction value obtained by tuning the first regression value, and a learning unit that generates an improvement model, which is another regression model for classifying data of the same type as the first regression model, by learning the first correction value in the first data set as correct data.

Advantages of the Invention

[0009] According to one aspect of the embodiment, further improvement in accuracy in a classification task can be achieved.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Embodiments for Carrying Out the Invention

[0011] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the drawings. In the following embodiments, the same parts are denoted by the same reference numerals, and redundant descriptions are omitted.

[0012] (1. Embodiment) (1-1. Outline of Prediction Processing According to the Embodiment) FIG. 1 is a diagram showing an outline of the prediction processing according to the embodiment. The prediction processing according to the embodiment is executed by a prediction device 100 (not shown in FIG. 1).

[0013] The prediction device 100 is an information processing device that generates a machine learning model for executing a task of classifying arbitrary data and executes classification and prediction processing using the generated model. For example, the prediction device 100 is a PC (Personal Computer), a tablet terminal, a smartphone, a cloud server, or the like.

[0014] As described above, according to the model generated by machine learning, classification processing in various tasks (such as data analysis) can be executed. Here, in the prediction processing according to the present disclosure, when performing classification, instead of simply using a classification model by machine learning, first a regression model is generated, and based on the predicted value of the regression obtained from the model, classification is performed under a certain condition.

[0015] With such a configuration, the prediction device 100 according to the present disclosure can improve the accuracy of the classification processing and improve the correct label in the learning data.

[0016] First, the outline of the prediction process according to the present disclosure will be described with reference to FIG. 1. First, the prediction device 100 generates a model that performs prior regression processing in order to generate a model that executes classification and prediction processing (hereinafter, the model that finally executes classification and prediction processing is referred to as an "improved regression model" for distinction).

[0017] In the example of FIG. 1, the prediction device 100 generates a first regression model 50 and a second regression model 60. Note that the first regression model 50 and the second regression model 60 are regression models pre-trained with learning data in which, for example, measured data and correct labels are combined.

[0018] In the embodiment, the first regression model 50 and the second regression model 60 are learned with learning data in which the body temperature data acquired from a device capable of continuously measuring the user's body temperature data daily and whether the user being measured is having a menstrual period are used as correct data. That is, the first regression model 50 and the second regression model 60 are models that classify and predict which day is the user's menstrual period when body temperature data (for example, body temperature data for one month measured at a fixed time every day) is acquired as input data.

[0019] Specifically, the first regression model 50 and the second regression model 60 output a regression value, which is a numerical value represented from 0 to 1, for the input body temperature data. The prediction device 100 predicts whether a certain day is the user's menstrual period according to whether such a regression value exceeds a threshold value (for example, "0.7") or not. Note that the threshold value is a value obtained by machine learning and is a value obtained by searching for where to set the boundary to most accurately determine the menstrual period. The threshold value exists for each model. As an example, if the regression value corresponding to the body temperature data of a certain day exceeds "0.7", the prediction device 100 determines that the day is the user's menstrual period. On the other hand, if the regression value corresponding to the body temperature data of a certain day is "0.7" or less, the prediction device 100 determines that the day is not the user's menstrual period.

[0020] The data used for training the first regression model 50 and the second regression model 60 is a combination of body temperature data and correct answer data. In this case, the correct answer data is a numerical value indicating the probability of whether the body temperature data on a certain day is the menstrual day or not, and is given as a numerical value from 0 to 1, for example. Such correct answer data may be data with a specific correct answer given as a label, or may be data with some value arbitrarily set by the data creator or the like. For example, when body temperature data is acquired from a user, the training data can be data to which correct answer data is added by recording whether or not the user's certain date was the menstrual day, etc.

[0021] When trained with the above training data, for example, the learned first regression model 50 returns a regression value corresponding to the input data when receiving the input of the input data (for example, body temperature data for a certain number of days newly acquired). The regression value is a regression value indicating the probability that the day corresponding to the input data, which is the body temperature data, is the menstrual day.

[0022] The prediction device 100 obtains a regression value as the output of the first regression model 50. Further, the prediction device 100 obtains a first data set 55 which is a combination of the output regression value and the correct answer data corresponding to the regression value. Although details will be described later, the correct answer data in the first data set 55 is originally a value that the first regression model 50 wants to output (calculate), and is obtained by a predetermined adjustment process (generally referred to as "tuning").

[0023] Similarly, the prediction device 100 inputs the same input data to the second regression model 60 and obtains a regression value as the output. Further, the prediction device 100 obtains a second data set 65 which is a combination of the output regression value and the correct answer data corresponding to the regression value. The correct answer data in the second data set 65 is originally a value that the second regression model 60 wants to output, similar to the first data set 55, and is obtained by tuning. Note that since the second regression model 60 has a different configuration (or a different learning method) from the first regression model 50, for example, it has a threshold value different from the threshold value set for the first regression model 50.

[0024] The dataset obtained as described above is used for training the improved regression model. That is, the prediction device 100 trains the pre-training improved regression model 70 using the first dataset 55 and the second dataset 65 for which sufficient amounts have been obtained as training data.

[0025] That is, the data used for training the pre-training improved regression model 70 is a combination of the regression values that are the output results of the first regression model 50 and the second regression model 60, and the correct data newly set by tuning based on the results. In this way, the prediction device 100 attempts not simply learning using a plurality of models (referred to as ensemble learning, etc.), but learning using the correct data newly set based on the predicted value of regression by the regression model. Thereby, the prediction device 100 can generate a model capable of obtaining a more improved regression value as compared with simply learning using a plurality of models, and thus can improve the classification accuracy.

[0026] Through training, the prediction device 100 obtains a trained improved regression model 80. The prediction device 100 inputs input data consisting of the user's body temperature data into the improved regression model 80 to obtain a more accurate classification result. In the example of FIG. 1, the prediction device 100 can predict or classify whether the user is having a menstrual period from the body temperature data of a certain day.

[0027] Next, the dataset will be described in detail with reference to FIGS. 2 and 3. FIG. 2 is a diagram showing an example of the first dataset 55 according to the embodiment.

[0028] Figure 2 shows the specific configuration of the dataset. The learning data 51 indicates the data used for learning the first regression model 50. As shown in Figure 2, the learning data 51 is configured such that the body temperature data and the correct answer data (label) are paired. As described above, the body temperature data is, for example, the body temperature data measured regularly every day by the user (as an example, the basal body temperature data measured by a women's thermometer or the like). Also, the correct answer data is, for example, a numerical value arbitrarily given by the data creator to indicate the probability of the physiological day, or a numerical value indicating whether or not the physiological day set by the user is the case.

[0029] The prediction device 100 learns the first regression model 52 before learning using the learning data 51. For example, the prediction device 100 sets the classification threshold α to "0.7" and generates a model for classifying whether or not it is the physiological day with such a threshold. Depending on the learning method, the threshold α is a numerical value obtained as a result of learning. Specifically, the prediction device 100 obtains that the threshold α is "0.7" by a method of deriving the threshold that enables classification with the highest accuracy.

[0030] The prediction device 100 obtains the (learned) first regression model 50 through learning. The prediction device 100 inputs the body temperature data obtained from the user into the first regression model 50 to obtain a classification result, that is, a regression value.

[0031] As shown in Figure 2, the regression value that is the output of the first regression model 50 indicates, by a numerical value from "0" to "1", whether or not the day corresponding to the body temperature data is the physiological day. For example, in Figure 2, the prediction device 100 obtains numerical values such as "0.031", "0.024", "0.123", "0.134" as the regression values for each date.

[0032] In contrast, the prediction device 100 further sets correct data. As described above, such correct data originally indicates the value that the first regression model 50 is to calculate. Such correct data is newly set by tuning based on the regression value output from the first regression model 50. For example, the prediction device 100 can use a known method such as Bayesian optimization as a tuning method. Specifically, the prediction device 100 can tune the correct data using so-called multi-objective Bayesian optimization that evaluates the final accuracy while changing the set correct data (correct label) to determine the optimal correct label. Note that the prediction device 100 can also adopt known methods such as grid search and random search as tuning methods. Note that the initial value of the above correct data is not particularly limited and may be, for example, a random number or the like. The prediction device 100 tunes to the correct data that is ideal for raising the final prediction accuracy by multi-objective Bayesian optimization or the like for such correct data.

[0033] As shown in FIG. 2, the prediction device 100 obtains numerical values such as "0.0", "0.0", "0.1", and "0.9" as the correct data corresponding to each regression value. Such a combination of the correct data and the regression value is an example of the first data set 55. Although not shown in FIG. 2, the first data set 55 may include the input data (such as body temperature data) input to the first regression model 50 that output the regression value.

[0034] Subsequently, the second data set 65 will be described with reference to FIG. 3. FIG. 3 is a diagram showing an example of the second data set 65 according to the embodiment.

[0035] As shown in FIG. 3, basically, the prediction device 100 creates the second data set 65 using the second regression model 60 in the same manner as creating the first data set 55 using the first regression model 50.

[0036] The pre-learning second regression model 62 is learned by learning data 61 in which body temperature data and correct answer data are combined, similar to the learning data 51. The learning data 51 and the learning data 61 may be data with the same content. As described above, since the pre-learning second regression model 62 differs in configuration (learning method, parameters of explanatory variables, model structure itself, etc.) from the pre-learning first regression model 52, the threshold β used for classification may have a different numerical value from the threshold α.

[0037] When the learned second regression model 60 is obtained by learning using the learning data 61, the prediction device 100 obtains a regression value that is the output of the second regression model 60, similar to FIG. 2. For example, in FIG. 3, the prediction device 100 obtains numerical values such as "0.033", "0.025", "0.046", and "0.031" as the regression values for each date.

[0038] Also, in the same method as FIG. 2, the prediction device 100 obtains numerical values such as "0.0", "0.0", "0.1", and "0.9" as the correct answer data corresponding to each regression value. Such a combination of the correct answer data and the regression value is an example of the second data set 65.

[0039] The prediction device 100 generates an improved regression model 80 by performing learning of the improved regression model using the first data set 55 and the second data set 65 shown in FIGS. 2 and 3. Note that the models for obtaining the data set for generating the improved regression model are not limited to the two models of the first regression model 50 and the second regression model 60, and more models may be prepared.

[0040] By performing the above processing, the prediction device 100 can generate an improved regression model 80 that can more sharply classify the days that are physiological days by suppressing the values (regression values) when the regression value is not a physiological day to be smaller.

[0041] Next, a practical example when the improved regression model 80 according to the embodiment is used will be shown with reference to FIG. 4. FIG. 4 is a diagram showing an application example of the prediction process according to the embodiment.

[0042] Specifically, FIG. 4 schematically shows, in the embodiment, the process of acquiring body temperature data from user 15, the process of generating a model based on the acquired body temperature data, or the flow when prediction processing is performed based on the acquired body temperature data.

[0043] The prediction processing according to the embodiment is executed by the prediction system 1 shown in FIG. 4. The prediction system 1 includes a prediction device 100 which is an example of the prediction device according to the present disclosure, a measurement device 10, and a user terminal 200. Each device included in the prediction system 1 can transmit and receive data to and from each other by wireless communication or the like.

[0044] The measurement device 10 is a device used for measuring basal body temperature. The measurement device 10 is a so-called wearable device, which is stored inside clothing such as the underwear of user 15 and measures the biological information of the user continuously or periodically (for example, at a fixed time every morning). In the example of FIG. 4, user 15 is a user who provides learning data (basal body temperature data, physiological days, etc.) used for generating the regression model according to the embodiment, or a user who obtains the prediction result of the next physiological day by transmitting the body temperature data of a predetermined number of days to the prediction device 100.

[0045] The measurement device 10 measures the basal body temperature of user 15 and holds the daily measured basal body temperature. Note that the basal body temperature is the body temperature in a resting state where only the minimum amount of energy necessary for maintaining life is consumed. Generally, it refers to the body temperature measured using a female thermometer while still in bed when waking up in the morning. The measurement device 10 may measure the basal body temperature itself, or may measure the skin temperature of the user. When measuring the skin temperature, the measurement device 10 may calculate an approximation value of the basal body temperature from the deep body temperature derived from the outside air temperature and the skin temperature, and handle the calculated data as measurement data (basal body temperature data). Further, the measurement device 10 may also measure data such as the heart rate of user 15 as biological information.

[0046] The user terminal 200 is a terminal device used by the user 15, such as a smartphone or a tablet terminal. An app for controlling the measurement device 10 is installed on the user terminal 200. By using the app operating on the user terminal 200, the user 15 can control the measurement device 10 or refer to the basal body temperature data measured by the measurement device 10. Further, the user terminal 200 acquires the basal body temperature data measured by the measurement device 10 and transmits the acquired basal body temperature data to the prediction device 100. Also, the user terminal 200 receives from the prediction device 100 the result classified or predicted by the prediction device 100. The user 15 can refer to the result predicted by the prediction device 100 via the user terminal 200. Note that, as described above, when the measurement device 10 measures the result calculated by arithmetic operation of data close to the basal body temperature using the skin temperature and outside air temperature data, the basal body temperature data may be the data collected of such calculation results. That is, the basal body temperature data is not necessarily limited to a series of data accumulating the basal body temperature, and may be data having a temperature change pattern equivalent to the basal body temperature.

[0047] Although not shown in FIG. 4, it is assumed that there are a large number of users who use the measurement device 10 in addition to the user 15. The prediction device 100 acquires basal body temperature data from these many users at any time, and generates a regression model and an improved regression model for creating a learning data set using the acquired data as learning data. That is, the user 15 is a general term for a large number of users who provide basal body temperature data as learning data.

[0048] As described above, the prediction device 100 predicts the user's menstrual date, that is, the menstrual cycle. The menstrual cycle has a periodicity (biphasic property) in which it enters a high temperature phase under the influence of hormones secreted with ovulation and then enters a low temperature phase after the start date of menstruation (menstruation). Therefore, by statistically processing the basal body temperature data, it is possible to predict the pattern of the menstrual cycle to some extent.

[0049] In the example shown in FIG. 4, user 15 obtains basal body temperature data every day by wearing measurement device 10 to go to bed. That is, measurement device 10 continuously measures the basal body temperature data of user 15 (step S1). Measurement device 10 sequentially or periodically transmits the measured basal body temperature data to user terminal 200 (step S2).

[0050] User terminal 200 sequentially or periodically transmits the basal body temperature data measured by measurement device 10 to prediction device 100 (step S3).

[0051] The basal body temperature data transmitted from user terminal 200 to prediction device 100 can be shown as a waveform by plotting the numerical values of the basal body temperature and the measurement date, for example, as graph 30 shown in FIG. 4. In addition, when user 15 can recognize the menstrual date, by registering the data indicating when the menstrual date is, the menstrual date can be included in the basal body temperature data.

[0052] Graph 30 shows the basal body temperature data for approximately one physiological cycle of user 15. Note that graph 30 is data having a biphasic tendency, with the first half showing a low temperature period and the second half showing a high temperature period.

[0053] Prediction device 100 accumulates the acquired basal body temperature data (measurement data by measurement device 10) as measurement data 90, and based on the acquired basal body temperature data, generates a regression model for classifying whether the date corresponding to the body temperature data is the menstrual date or not (step S4). In other words, prediction device 100 performs machine learning by AI based on the learning data composed of the body temperature data and the correct label regarding the menstrual date, thereby generating a regression model. Specifically, prediction device 100 generates first regression model 50 and second regression model 60.

[0054] Furthermore, the prediction device 100 inputs the measurement data acquired from the user 15 into the first regression model 50 and the second regression model 60, and generates an improved regression model 80 based on a data set formed by the combination of the regression values output from each model and the correct data.

[0055] After that, when the prediction device 100 acquires measurement data from the user 15, it performs a process of classifying whether the date corresponding to any body temperature data is the menstrual period or not using the improved regression model 80. Furthermore, the prediction device 100 may return a prediction result to the user 15, such as whether the menstrual period will arrive in a few days, based on the trend of the body temperature data continuously acquired from the user 15 every day.

[0056] In this way, by performing classification and prediction processing using the improved regression model, the prediction device 100 can provide more accurate classification and prediction results compared to the conventional regression model.

[0057] (1-2. Configuration of the prediction device according to the embodiment) Next, the configuration of the prediction device 100 that executes the prediction process according to the embodiment will be described. FIG. 5 is a diagram showing a configuration example of the prediction device 100 according to the embodiment.

[0058] As shown in FIG. 5, the prediction device 100 includes a communication unit 110, a storage unit 120, and a control unit 130. Note that the prediction device 100 may have an input unit (for example, a keyboard, a mouse, etc.) that receives various operations from an administrator or the like who manages the prediction device 100, and a display unit (for example, a liquid crystal display, etc.) for displaying various information.

[0059] The communication unit 110 is realized by, for example, a Network Interface Controller or the like. The communication unit 110 is connected to the network N (for example, the Internet) by wire or wirelessly, and transmits and receives information to and from the measurement device 10, the user terminal 200, etc. via the network N. For example, the communication unit 110 may transmit and receive information using a communication standard or communication technology such as Wi-Fi (registered trademark), SIM (Subscriber Identity Module), LPWA (Low Power Wide Area).

[0060] The storage unit 120 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. The storage unit 120 has a measurement data storage unit 121 and a model storage unit 122.

[0061] The measurement data storage unit 121 stores measurement data and the like acquired from a user wearing the measurement device 10. That is, the measurement data storage unit 121 stores learning data (body temperature data and correct answer data) used for learning the regression model, and a data set (body temperature data, regression value, and correct answer data tuned based on the regression value) used for learning the improved regression model.

[0062] The model storage unit 122 stores the regression model and the improved regression model used by the prediction device 100 for processing. Note that the model storage unit 122 may store a plurality of regression models such as regression models with different lengths of the number of days of unit data and regression models with different configurations.

[0063] The control unit 130 is realized, for example, by a program stored inside the prediction device 100 being executed with a RAM or the like as a work area by a CPU (Central Processing Unit), an MPU (Micro Processing Unit), a GPU (Graphics Processing Unit), or the like. Further, the control unit 130 is a controller and is realized, for example, by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0064] As shown in FIG. 5, the control unit 130 includes an acquisition unit 131, a learning unit 132, a prediction unit 133, and a provision unit 134.

[0065] The acquisition unit 131 acquires various types of information. For example, the acquisition unit 131 acquires various types of data related to the classification task. Specifically, the acquisition unit 131 acquires body temperature data during the physiological cycle from a user who uses the measurement device 10. More specifically, the acquisition unit 131 controls a program (application) installed in the user terminal 200 and acquires body temperature data from the user terminal 200 at the timing when the body temperature data is acquired by the user terminal 200, at the timing of daily fixed times, or the like.

[0066] Further, the acquisition unit 131 acquires learning data for generating a regression model and an improved regression model. For example, the acquisition unit 131 acquires, as learning data for the regression model, a combination of arbitrary data that is the target of the classification task and correct answer data for the data. The correct answer data for such learning data may be, for example, a numerical value manually set by an administrator or a numerical value set by a data provider.

[0067] Further, the acquisition unit 131 acquires a data set composed of a combination of a regression value output when arbitrary data is input to the first regression model 50 or the like and a correction value (tuned correct data) obtained by tuning the first regression value. For example, as tuning, the acquisition unit 131 acquires correct data by any one of grid search, random search, and Bayesian optimization. Such a data set is used as learning data for the improved regression model 80.

[0068] As shown in FIGS. 2 and 3, the acquisition unit 131 may acquire a data set from a plurality of regression models (the first regression model 50 and the second regression model 60 in the embodiment) instead of a single regression model. Thereby, the acquisition unit 131 can increase the quantity of the data set and improve the accuracy of the correct data in the data set.

[0069] Also, the acquisition unit 131 acquires input data to be input to the improved regression model 80 according to, for example, a user's request. Specifically, the acquisition unit 131 acquires the user's body temperature data for a predetermined number of days as the input data. The prediction device 100 classifies such input data, performs prediction processing according to the user's request, and returns the classification result or the prediction result to the user.

[0070] When the prediction device 100 handles body temperature data, the acquisition unit 131 may acquire the body temperature data continuously measured by a wearable device (measurement device 10 in the example of FIG. 4) worn by the user. Thereby, the acquisition unit 131 can acquire data with less error and high reliability. Note that the acquisition unit 131 may acquire not only body temperature data but also all body information that is assumed to affect the physiological cycle, such as the age, height, and weight of each user. The acquisition unit 131 acquires these pieces of information by receiving input from the user, for example, via user registration in an application. Further, when the measurement device 10 can measure biological information other than body temperature, such as a heart rate, the acquisition unit 131 may acquire the biological information together with the body temperature data. Further, when acquiring time-series data such as body temperature data continuously acquired daily, even if there are missing parts due to the user forgetting to measure or measurement errors, the acquisition unit 131 may complement the data by a known interpolation technique such as interpolating the missing parts with the data before and after. The acquisition unit 131 stores the acquired various data in the measurement data storage unit 121.

[0071] The learning unit 132 generates an improved regression model for classifying data of the same type as the model (such as the first regression model 50) that output a regression value in the previous stage by learning the correction value in the dataset as correct data.

[0072] As described above, the learning unit 132 generates the improved regression model 80 using the correct data tuned based on the regression values output from each of the plurality of regression models.

[0073] For example, when the data handled in the classification task is body temperature data and the classification task is "whether it is the physiological day or not", the learning unit 132 generates an improved regression model 80 that outputs a regression value indicating whether the date corresponding to the body temperature data is the physiological day or not. The learning unit 132 stores the generated improved regression model 80 in the model storage unit 122.

[0074] The prediction unit 133 inputs the input data acquired by the acquisition unit 131 into the improved regression model 80, and outputs a classification result by the improved regression model 80 or a prediction result based on the classification result.

[0075] For example, the prediction unit 133 inputs the body temperature data of the user for a predetermined number of days into the improved regression model 80 that is trained to classify whether the day corresponding to the predetermined body temperature data is the user's menstrual day, and outputs the classification result for each of the predetermined number of days. Further, the prediction unit 133 may output a prediction result based on the classification result. As an example, the prediction unit 133 outputs a prediction result such as the arrival of the user's menstrual day one or two days later based on the trend of the regression values for each date. The prediction unit 133 may utilize known machine learning techniques and the like in such prediction processing.

[0076] The providing unit 134 provides the user with the classification result output by the prediction unit 133 or the prediction result based on the classification result.

[0077] For example, the providing unit 134 provides the user with a prediction result of the menstrual day and the like based on the classification result (whether each date is the menstrual day) by the prediction unit 133. For example, the providing unit 134 provides the prediction result to the user by displaying the prediction result on the user terminal 200. At this time, when a sign such as the approach of the menstrual day is observed in the prediction result, the providing unit 134 may perform a more emphasized display (notification) to the user than usual. Further, the providing unit 134 may transmit advice to the user based on the result predicted by the prediction unit 133. For example, the providing unit 134 may transmit advice regarding the user's physical condition and pregnancy possibility estimated from the prediction result.

[0078] The providing unit 134 may send different advice for each user. For example, the providing unit 134 refers to past history (such as registration by the user), and obtains information that the user tends to have a poor physical condition several days before the start of menstruation. In this case, the providing unit 134 may send advice to the user indicating the date several days before the predicted start date of menstruation and suggesting that there is a possibility of poor physical condition during that period. On the other hand, the providing unit 134 may take the measure of not sending such advice to users who do not tend to have a poor physical condition several days before the start of menstruation. Such processing for each user can be automated by machine learning the behavior history of each user, etc.

[0079] (1-3. Procedure of the process according to the embodiment) The procedure of the processing of the above-described prediction device 100 will be described with reference to FIGS. 6 to 8. FIG. 6 is a flowchart showing the procedure of the regression model generation processing according to the embodiment.

[0080] As shown in FIG. 6, the prediction device 100 acquires body temperature data used for the learning process from the user 15 (step S101).

[0081] Subsequently, the prediction device 100 creates teacher data (learning data) from the acquired body temperature data (for example, body temperature data for one or more cycles measured from the day including the previous physiological day to the day including the next physiological day) and correct data (a value indicating the accuracy of whether it is a physiological day) combined with the body temperature data (step S102).

[0082] Using the learning data created in step S102, the prediction device 100 generates a regression model (such as the first regression model 50 and the second regression model 60) for creating a data set for generating an improved regression model (step S103).

[0083] Subsequently, the generation process of the improved regression model 80 will be described with reference to FIG. 7. FIG. 7 is a flowchart showing the procedure of the improved regression model generation processing according to the embodiment.

[0084] The prediction device 100 inputs input data such as body temperature data into the regression model generated in step S103 (step S201). The prediction device 100 obtains a regression value as the output of the regression model (step S202).

[0085] Subsequently, the prediction device 100 tunes the correct data to be combined with the regression value based on the regression value (step S203). Then, the prediction device 100 creates a data set in which the regression value and the correct data are combined (step S204).

[0086] Subsequently, the prediction device 100 determines whether a data set necessary for generating an improved regression model has been accumulated (step S205). If the data set is not sufficient (step S205; No), the prediction device 100 creates a data set further.

[0087] On the other hand, when a data set sufficient for learning has been accumulated (step S205; Yes), the prediction device 100 learns an improved regression model using the data sets obtained from each regression model, and generates an improved regression model (step S206).

[0088] Subsequently, with reference to FIG. 8, the prediction or classification process by the prediction device 100 will be described. FIG. 8 is a flowchart showing the procedure of the classification and prediction process according to the embodiment.

[0089] The prediction device 100 acquires a certain amount of body temperature data (for example, body temperature data measured continuously for one week or one month) from a user who is a person to be predicted (step S301).

[0090] The prediction device 100 inputs the acquired body temperature data into the improved regression model (step S302). Then, the prediction device 100 classifies whether it is the menstrual day for each date based on the regression value obtained from the improved regression model, or predicts the menstrual day based on the classification result (step S303).

[0091] Subsequently, the prediction device 100 provides the user with the processing result and advice (step S304).

[0092] (2. Modification of the Embodiment) In the above embodiment, an example in which the user terminal 200 according to the present disclosure acquires the body temperature data measured by the measurement device 10 is shown. However, the user terminal 200 does not necessarily have to acquire the body temperature data from the measurement device 10. For example, the user terminal 200 may directly receive an input from the user regarding the body temperature data acquired by the user using a general thermometer, and handle the received data as input data or learning data.

[0093] Also, in the embodiment, an example in which the user terminal 200 transmits measurement data to the prediction device 100 and the prediction device 100 classifies the menstrual cycle is shown. However, when the user terminal 200 has a regression model, the user terminal 200 may predict the user's menstrual cycle based on the body temperature data measured by the measurement device 10. That is, the processing executed by the prediction device 100 in the embodiment may be executed by the user terminal 200.

[0094] Also, in the embodiment, an example in which the prediction device 100 has the storage unit 120 and holds the measurement data and the regression model is shown, but these pieces of information may be held by an external storage device other than the prediction device 100, a storage server on a network, or the like.

[0095] In addition, in the embodiment, an example in which the prediction device 100 handles data including periodicity and time series such as determination of the user's menstrual cycle was shown. However, the prediction device 100 may handle any information as long as it is data that can be handled as a classification task. In this case, the first regression model 50 etc. for creating a data set and the improved regression model 80 for executing a classification task shall handle data with the same time series, structure, etc. (for example, body temperature data is stored for each consecutive date). This is because the correct data needs to be common. Note that even if the improved regression model 80 is learned from time series data, when the prediction device 100 executes classification of, for example, a single piece of data, it does not need to input time series data and can also input a single piece of data to be classified.

[0096] (3. Other Embodiments) The processing according to the above-described embodiment may be implemented in various different forms other than the above embodiment.

[0097] For example, among the respective processes described in the above embodiment, all or part of the processes described as being automatically performed can also be performed manually, or all or part of the processes described as being performed manually can be automatically performed by a known method. In addition, regarding the processing procedures, specific names, and information including various data and parameters shown in the above document and drawings, they can be arbitrarily changed unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

[0098] Also, each component of each device shown in the drawings is conceptually functional and does not necessarily have to be physically configured as shown in the drawings. That is, the specific form of distribution and integration of each device is not limited to that shown in the drawings, and all or part of it can be functionally or physically distributed and integrated in any unit according to various loads, usage situations, etc.

[0099] Also, the above-described embodiments and modification examples can be appropriately combined within a range that does not conflict with the processing content.

[0100] Also, the effects described in this specification are merely examples and are not limiting; there may be other effects.

[0101] (4. Effects of the prediction device according to the present disclosure) As described above, the prediction device (prediction device 100 in the embodiment) according to the present disclosure includes an acquisition unit (acquisition unit 131 in the embodiment) and a learning unit (learning unit 132 in the embodiment). The acquisition unit acquires a first data set (first data set 55 etc. in the embodiment) composed of a combination of a first regression value output when arbitrary data is input to a first regression model (first regression model 50 etc. in the embodiment) and a first correction value (correct data) obtained by tuning the first regression value. The learning unit generates an improved model (improved regression model 80 in the embodiment), which is another regression model for classifying data of the same type as the first regression model, by learning the first correction value in the first data set as correct data. Note that the acquisition unit acquires the first correction value by any one of grid search, random search, and Bayesian optimization as tuning.

[0102] In this way, the prediction device according to the present disclosure, as a pre-stage process, learns a regression model (improved model) that performs a post-stage process using, as correct data, a correction value tuned based on the regression value output from the regression model. As a result, the prediction device can obtain a regression model learned with more accurate correct data, and thus can achieve a further improvement in accuracy in the classification task.

[0103] Also, the acquisition unit acquires a second data set (second data set 65 etc. in the embodiment) composed of a combination of a second regression value output when arbitrary data is input to a second regression model (second regression model 60 etc. in the embodiment), which is a regression model different from the first regression model, and a second correction value obtained by tuning the second regression value. The learning unit generates an improved model by learning the second correction value as correct data in addition to the first correction value.

[0104] In this way, the prediction device can obtain a dataset for learning from a plurality of regression models in the preprocessing stage, so that the accuracy of the subsequent improvement model can be further improved.

[0105] The prediction device may further include a prediction unit (prediction unit 133 in the embodiment) and a providing unit (providing unit 134 in the embodiment). The acquisition unit acquires input data for inputting to the improvement model from the user. The prediction unit inputs the acquired input data into the improvement model and outputs a classification result by the improvement model or a prediction result based on the classification result. The providing unit provides the user with the classification result output by the prediction unit or the prediction result based on the classification result.

[0106] In this way, the prediction device can provide the user with a highly accurate classification result by performing classification processing and prediction processing.

[0107] In addition, the acquisition unit acquires the body temperature data of the user for a predetermined number of days as input data. The prediction unit inputs the body temperature data of the user for a predetermined number of days into an improvement model learned to classify whether the day corresponding to the predetermined body temperature data is the user's physiological day, and outputs the classification result for each of the predetermined number of days. The providing unit provides the user with the classification result or prediction result of the physiological day based on the classification result by the prediction unit.

[0108] For example, the acquisition unit acquires the body temperature data continuously measured by a wearable device (measurement device 10 in the embodiment) worn by the user.

[0109] In this way, the prediction device can accurately classify time series data such as body temperature data for determining the physiological day, so that useful information with high accuracy can be provided to the user.

[0110] (5. Hardware Configuration) The information devices such as the prediction device 100 and the user terminal 200 according to the above-described embodiments are realized by a computer 1000 configured as shown in FIG. 9, for example. Hereinafter, the prediction device 100 according to the embodiment will be described as an example. FIG. 9 is a hardware configuration diagram showing an example of a computer 1000 that realizes the functions of the prediction device 100. The computer 1000 includes a CPU 1100, a RAM 1200, a ROM (Read Only Memory) 1300, an HDD (Hard Disk Drive) 1400, a communication interface 1500, and an input / output interface 1600. Each part of the computer 1000 is connected by a bus 1050.

[0111] The CPU 1100 operates based on a program stored in the ROM 1300 or the HDD 1400 and controls each part. For example, the CPU 1100 expands a program stored in the ROM 1300 or the HDD 1400 into the RAM 1200 and executes processing corresponding to various programs.

[0112] The ROM 1300 stores a boot program such as a BIOS (Basic Input Output System) executed by the CPU 1100 when the computer 1000 is started up, a program dependent on the hardware of the computer 1000, and the like.

[0113] The HDD 1400 is a computer-readable recording medium that non-temporarily records a program executed by the CPU 1100 and data used by such a program. Specifically, the HDD 1400 is a recording medium that records a program for executing the prediction process according to the present disclosure, which is an example of the program data 1450.

[0114] The communication interface 1500 is an interface for the computer 1000 to connect to an external network 1550 (for example, the Internet). For example, the CPU 1100 receives data from other devices or transmits data generated by the CPU 1100 to other devices via the communication interface 1500.

[0115] The input / output interface 1600 is an interface for connecting the input / output device 1650 and the computer 1000. For example, the CPU 1100 receives data from input devices such as a keyboard and a mouse via the input / output interface 1600. Also, the CPU 1100 transmits data to output devices such as a display, a speaker, and a printer via the input / output interface 1600. Further, the input / output interface 1600 may function as a media interface for reading a program or the like recorded on a predetermined recording medium (media). The media is, for example, an optical recording medium such as a DVD (Digital Versatile Disc) or a PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory or the like.

[0116] For example, when the computer 1000 functions as the prediction device 100 according to the embodiment, the CPU 1100 of the computer 1000 realizes functions such as the control unit 130 by executing the prediction processing program loaded on the RAM 1200. Also, the HDD 1400 stores a program for executing the prediction processing according to the present disclosure and data in the storage unit 120. Note that the CPU 1100 reads and executes the program data 1450 from the HDD 1400, but as another example, these programs may be acquired from other devices via the external network 1550.

[0117] As described above, the embodiments of the present application have been described in detail with reference to the drawings, but these are examples, and the present invention can be implemented in other forms in which various modifications and improvements are made based on the knowledge of those skilled in the art, including the aspects described in the column of the disclosure of the invention.

Description of Reference Numerals

[0118] 1 Prediction system 10 Measuring device 50 First regression model 60 Second regression model 80 Improved regression model 100 Prediction device 110 Communication unit 120 Memory unit 121 Measurement data memory unit 122 Model memory unit 130 Control unit 131 Acquisition unit 132 Learning unit 133 Prediction unit 134 Provision unit 200 User terminal

Claims

1. An acquisition unit that acquires a first dataset composed of a combination of a first regression value output when arbitrary data is input to a first regression model and a first correction value obtained by tuning the first regression value; A learning unit that generates an improved model, which is another regression model for classifying data of the same type as the first regression model, by learning the first correction value in the first dataset as correct data; A prediction device, characterized by comprising:

2. The acquisition unit: Acquires a second dataset composed of a combination of a second regression value output when arbitrary data is input to a second regression model, which is a regression model different from the first regression model, and a second correction value obtained by tuning the second regression value; The learning unit: Generates the improved model by learning the second correction value as correct data in addition to the first correction value; The prediction device according to claim 1, characterized by the above.

3. The acquisition unit: Acquires input data for inputting to the improved model from a user; The prediction device: Inputs the acquired input data into the improved model, and outputs a classification result by the improved model or a prediction result based on the classification result; A providing unit that provides the user with the classification result or the prediction result based on the classification result output by the prediction unit; The prediction device according to claim 1, further characterized by comprising:

4. The acquisition unit: Acquires the user's body temperature data for a predetermined number of days as the input data; The prediction unit: Inputs the user's body temperature data for a predetermined number of days into the improved model that is learned to classify whether the day corresponding to the predetermined body temperature data is the user's menstrual day, and outputs a classification result for each of the predetermined number of days; The providing unit: Provides the user with a classification result or a prediction result of the menstrual day based on the classification result by the prediction unit; The prediction device according to claim 3, characterized by the above.

5. The acquisition unit: Acquires body temperature data continuously measured by a wearable device worn by the user; The prediction device according to claim 4, characterized by the above.

6. The acquisition unit: As the tuning, acquires the first correction value by any one of the methods of grid search, random search, and Bayesian optimization; The prediction device according to claim 1, characterized in that...

7. The computer obtains a first data set consisting of a combination of a first regression value output when arbitrary data is input to a first regression model and a first correction value obtained by tuning the first regression value, and generates an improved model, which is another regression model for classifying data of the same type as the first regression model, by learning using the first correction value in the first data set as correct data. A prediction method, characterized by including the above.

8. An acquisition unit that acquires a first data set consisting of a combination of a first regression value output when arbitrary data is input to a first regression model and a first correction value obtained by tuning the first regression value, and a learning unit that generates an improved model, which is another regression model for classifying data of the same type as the first regression model, by learning using the first correction value in the first data set as correct data. A prediction program, characterized by causing a prediction device provided with the above to function.

Images