Information processing device, information processing method, and program

Abstract

This information processing device (1) comprises: an extraction unit (11) that extracts information of at least a portion of values from a result of regression analysis of a data set that is used for learning; and an output unit (12) that outputs information of a step in which data corresponding to the information of the portion of values extracted by the extraction unit (11) was acquired from among steps in which the data set was acquired.

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Regression analysis is a method for predicting the deterioration of an analysis target such as a system or a device. Regression analysis analyzes the relationship between one or more explanatory variables and one target variable in a dataset, and uses the resulting relationship to predict unknown data. For example, the anomaly detection method described in Patent Document 1 acquires data from a plurality of sensors, divides the trajectory in the data space into a plurality of clusters based on the temporal change of the data, models the divided cluster group by regression analysis, and calculates an outlier as an anomaly candidate.

Prior Art Documents

Patent Documents

[0003] [[ID=2�]]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The conventional technique described in Patent Document 1 has a problem that the interpretability of the regression-analyzed data is low. For example, when using the regression-analyzed data for learning of deterioration detection, the variation in the learning data can cause over-detection. If the interpretability of the data is low, although it is normal data, the source of the data with variation is unknown, and it is difficult to distinguish the data with variation. Therefore, there is a possibility that learning data containing a lot of variation is used for learning.

[0005] The present disclosure solves the above problems, and an object thereof is to obtain an information processing apparatus that can provide information for improving the interpretability of a regression-analyzed dataset. [Means for solving the problem]

[0006] The information processing device relating to this disclosure includes an extraction unit that extracts information on at least some values ​​from the results of regression analysis of a dataset used for learning, and an output unit that outputs information on the process in which data corresponding to the information on some values ​​extracted by the extraction unit was acquired. [Effects of the Invention]

[0007] According to the information processing device described herein, it is possible to provide information that improves the interpretability of a regression-analyzed dataset. [Brief explanation of the drawing]

[0008] [Figure 1] This is a block diagram showing an example configuration of an information processing device according to Embodiment 1. [Figure 2] This figure shows the results of regression analysis performed on the dataset. [Figure 3] Figures 3A, 3B, and 3C show the results of regression analysis performed on various datasets. [Figure 4] This is a flowchart showing the information processing method according to Embodiment 1. [Figure 5] This diagram shows an overview of the outlier extraction process. [Figure 6] This figure shows the results of regression analysis on a dataset after outliers have been removed. [Figure 7] Figures 7A, 7B, and 7C show an overview of the interval extraction process. [Figure 8] This is a block diagram showing an example of a hardware configuration that realizes the functions of the information processing device according to Embodiment 1. [Figure 9] This is a block diagram showing an example configuration of an information processing device according to Embodiment 2. [Figure 10] This is a block diagram showing an example configuration of an information processing device according to Embodiment 3. [Figure 11]It is a flowchart showing the information processing method according to Embodiment 3. [Figure 12] It is a block diagram showing a configuration example of the information processing apparatus according to Embodiment 4. [Figure 13] It is a flowchart showing the information processing method according to Embodiment 4. [Figure 14] It is a block diagram showing a configuration example of the information processing apparatus according to Embodiment 5. [Figure 15] It is a flowchart showing the information processing method according to Embodiment 5. [Figure 16] It is a block diagram showing the configuration of Modification 1 of the information processing apparatus according to Embodiment 5. [Figure 17] It is a flowchart showing Modification 2 of the information processing method according to Embodiment 5. [Figure 18] It is a diagram showing an outline of a process for presenting data of deterioration candidates. [Figure 19] It is a block diagram showing a configuration example of the information processing apparatus according to Embodiment 6. [Figure 20] It is a flowchart showing the information processing method according to Embodiment 6.

Modes for Carrying Out the Invention

[0009] Embodiment 1. (Basic Configuration of Information Processing Apparatus) FIG. 1 is a block diagram showing a configuration example of an information processing apparatus 1 according to Embodiment 1. In FIG. 1, the information processing apparatus 1 extracts information on at least some values from the result of regression analysis of a data set used for learning (hereinafter referred to as the regression analysis result), and outputs information on the process in which data corresponding to the information on the extracted some values is acquired among each process in which the data set is acquired. Some values extracted from the regression analysis result are the values of analysis points obtained by performing regression analysis on the data. The data corresponding to the information on some values is the data before regression analysis corresponding to the analysis points that are some values.

[0010] For example, the information processing apparatus 1 acquires the regression analysis result from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis apparatus that performs a regression analysis on a preset data set. Alternatively, the information processing apparatus 1 may acquire the regression analysis result stored in the storage unit included in itself and generate process information based on the acquired information.

[0011] (Process information) The process information is information indicating a process in which data corresponding to information of some values extracted from the regression analysis result is acquired. The "process" is, for example, a process in which data corresponding to the information of some values is acquired among the processes represented by parameters characterizing repetitions such as work procedures or operations of regression analysis objects.

[0012] (Acquisition of regression analysis result) The regression analysis result includes, for example, information indicating attributes and characteristics of the data set, information indicating an outline of the regression model, information indicating coefficients and their meanings, information indicating the goodness of fit of the regression model, information indicating statistical significance, information indicating the reliability of prediction, and information indicating constraints and preconditions of the regression model. The information indicating attributes and characteristics of the data set includes, for example, parameters serving as explanatory variables and target variables, and information indicating the acquisition source of the data set. The information indicating an outline of the regression model includes, for example, whether the regression model used is simple regression analysis or multiple regression analysis, an equation indicating the regression model, and explanations of parameters. The information indicating coefficients and their meanings is information indicating coefficients of each explanatory variable and their meanings, and for example, indicates how each explanatory variable has affected the target variable. The information indicating the goodness of fit of the regression model includes, for example, evaluation information on how well the regression model fits the data based on the coefficient of determination or the distribution of residuals. The information indicating statistical significance includes, for example, evaluation information on whether each coefficient is statistically significant. Information indicating the reliability of a prediction includes, for example, information showing the prediction interval or confidence interval. Information indicating the constraints and assumptions of a regression model includes, for example, assumptions such as linearity, independence, and uniform variance, as well as information regarding the scope of application of the regression analysis.

[0013] For example, let's explain how to obtain regression analysis results detected in the deterioration detection of a railway motor. A vibration sensor attached to a railway vehicle detects vibrations generated in the railway motor while the railway vehicle is in operation. The vibration data sequentially detected by the vibration sensor becomes the dataset to be subjected to regression analysis, and is output to a regression analysis device (not shown in Figure 1). Furthermore, the regression analysis device and the information processing device 1 output operation data synchronized with the vibration data. The operation data is data indicating the driving operations of a railway vehicle powered by a railway motor. Furthermore, synchronization with vibration data refers to temporal synchronization; for example, operation data representing driving operations performed at a certain time is synchronized with vibration data generated at that time. The regression analysis device performs regression analysis using operational data as explanatory variables and vibration data as the dependent variable. The results of the regression analysis include, for example, the parameters of the explanatory and dependent variables, and information indicating the confidence interval. The regression analysis results generated by the regression analysis device are output from the regression analysis device to the information processing device 1. The dataset used for regression analysis could be, for example, data obtained by sequentially detecting the rotational speed of a railway motor, or data obtained by sequentially detecting the position, speed, acceleration, or torque of an industrial robot's axis. It could also be data obtained by sequentially detecting vibrations generated in a motor of an elevator, or data obtained by sequentially detecting the power used for a single opening and closing of a platform door installed on a station platform.

[0014] (Variability of regression analysis results) Figure 2 shows the regression analysis results, specifically the results detected in the degradation detection of railway motors. In Figure 2, the horizontal axis represents the explanatory variable, which is the operation data of the railway vehicle, and the vertical axis represents the dependent variable, which is the vibration data. C1 is the confidence interval. Point D is a point obtained by regression analysis of vibration data when the railway motor is in a normal state (hereinafter referred to as normal data), and it falls within the confidence interval C1. The predicted value E is the predicted value of the dependent variable obtained by regression analysis. Point F is a point obtained by regression analysis of normal data that falls outside the confidence interval C1. Here, the existence of points corresponding to normal data that fall outside the confidence interval C1 is called the variability of the normal data. In general, if degradation detection is trained using normal data with variability as training data, it leads to a decrease in the detection accuracy of the trained model obtained by this training. In particular, detection using regression models loses time-series information, making it difficult to adequately analyze outliers and increasing the likelihood of over-detection.

[0015] Therefore, the information processing device 1 extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was obtained among the processes from which the dataset was acquired. By referring to the process information, it is possible to identify the process from which data corresponding to the information on some values ​​extracted from the regression analysis results was obtained. Based on the identified process, even if there is variability in the data corresponding to some values, it can be distinguished from other data.

[0016] For example, the information processing device 1 presents operation data, which is an explanatory variable that yields points F that fall outside the confidence interval C1 among the points obtained by regression analysis of normal data, as process information. By referring to this process information, data with variability can be excluded from the training data. Furthermore, for example, the information processing device 1 can automatically extract tasks or driving sections where data variability occurs continuously, using parameters that characterize the repetition of driving patterns (e.g., kilometers) and a distance criterion specified by the user, and display this information on the display device 2. By referring to the displayed information, data with variability can be removed from the training data.

[0017] Display device 2 displays information output from information processing device 1. For example, display device 2 retrieves and displays information from information processing device 1 for the process in which data corresponding to some of the values ​​extracted by the extraction unit was acquired from the process in which the dataset was acquired. By referring to the process information displayed on display device 2, it is possible to identify the process in which data corresponding to some of the values ​​extracted from the data of the regression analysis results was acquired. Based on the identified process, even if there is variation in the data corresponding to some of the values, it can be distinguished from other data. Note that display device 2 may be a display unit provided by information processing device 1, or it may be a display unit provided by an external device that is communicated with information processing device 1 via a network. For example, display device 2 may be an LCD (Liquid Crystal Display) or an organic EL (Electroluminescence).

[0018] Input device 3 accepts input operations from outside the information processing device 1. If the information processing device 1 is a smartphone or tablet terminal, input device 3 is, for example, a touch panel integrated with the screen of the display device 2. If the information processing device 1 is a PC, input device 3 is, for example, a mouse or keyboard.

[0019] The information processing device 1 comprises an extraction unit 11 and an output unit 12. For example, the information processing device 1 is implemented by a computer. The computer's memory stores programs that constitute information processing applications for realizing the functions of the extraction unit 11 and the output unit 12. The computer's processor executes the information processing applications read from the memory, thereby realizing the functions of the extraction unit 11 and the output unit 12.

[0020] (Extraction part) The extraction unit 11 extracts information on at least some of the values ​​from the results of regression analysis performed on the dataset used for training. For example, if the information processing device 1 includes a communication unit (not shown in Figure 1) and can communicate with a regression analysis device via a network, the extraction unit 11 controls the communication unit to obtain information from the regression analysis device via the network indicating the results of regression analysis on the dataset. Furthermore, for example, if the information processing device 1 includes a storage unit (not shown in Figure 1) and the storage unit stores information indicating the results of regression analysis of the dataset, the extraction unit 11 may read and obtain the information indicating the results of regression analysis of the dataset from the storage unit. In this case, the information processing device 1 does not need to include a communication unit.

[0021] (Output section) The output unit 12 outputs information about the process from which data was acquired that corresponds to some of the values ​​extracted by the extraction unit 11 from among the processes from which the dataset was acquired. For example, consider a dataset in which vibration data, detected from vibrations occurring in the railway motor of a railway vehicle while the vehicle is repeatedly operating on a certain line, is subjected to regression analysis. The extraction unit 11 extracts outliers from the confidence interval from the results of the regression analysis of the dataset. Outliers are the values ​​extracted from the regression analysis results by the extraction unit 11. The output unit 12 uses the mileage, which is a parameter that characterizes the repetition of the operating pattern, and a distance criterion specified by the user, to output information about the operating section in which the points of the outliers extracted by the extraction unit 11 occur consecutively and data corresponding to the outliers are acquired, within the operating section represented by the mileage. This information about the operating section is the information for the above process. Here, the distance criterion is that anything less than 0.5 km is considered close, and anything greater than that is considered far.

[0022] For example, the output unit 12 calculates an invariant parameter Inv(t) for each repeated operation of the railway vehicle based on control information that controls the operation of the railway vehicle. Here, the parameter Inv(t) satisfies the condition (1) that it increases monotonically with respect to time t, and the condition (2) that there exists at least one I that is greater than or equal to 0, and that the range from 0 to I is repeated for each lap. The above-mentioned kilometer mile is the parameter Inv(t). The parameter Inv(t) is not limited to kilometers, and various parameters can be used depending on the subject of the regression analysis. For example, if the subject of the regression analysis is a series of tasks performed by an industrial robot as shown in Figure 3B (described later), the task progress, which indicates how far the series of tasks has been completed, can be used as the parameter Inv(t). Also, if the subject of the regression analysis is the opening and closing of platform screen doors as shown in Figure 3C (described later), the power used when the platform screen doors open and close once can be used as the parameter Inv(t). The output unit 12 may acquire control information of the object to be analyzed for regression (for example, a railway vehicle), calculate a cumulative function of all parameters Inv(t) based on the acquired control information, and then automatically calculate the parameter Inv(t) representing the kilometer distance by searching for the parameter that satisfies the above conditions (1) and (2) among the parameters for which the cumulative function has been calculated. Furthermore, the parameter Inv(t) may be set by the user to the output unit 12 using the input device 3.

[0023] The dataset may be a dataset obtained by repeatedly performing the process. Figures 3A, 3B, and 3C show outliers extracted from the regression analysis results of various datasets obtained by repeatedly performing the process. Figure 3A shows the relationship between outliers corresponding to vibration data of railway motors detected when a railway vehicle repeatedly circles a circular track, and the kilometer mileage representing the track the railway vehicle travels, over time. In Figure 3A, if the cause of outliers lies in a part of the track, the outliers for each lap will appear consecutively at the kilometer mileage corresponding to that part. Figure 3B shows the relationship between outliers in robot axis vibration data detected when an industrial robot repeatedly performs a series of tasks, and the task progress, which represents the progress of the series of tasks, over time. Similarly in Figure 3B, if there is a cause for outliers in a part of the series of tasks, the outliers in each cycle will appear consecutively at the task progress corresponding to that part. Figure 3C shows the relationship between outliers in power consumption data when platform doors open and close in response to the arrival and departure of railway vehicles at the platform, and the mileage representing the railway line on which the railway vehicles travel, over time. In Figure 3C, if there is a cause for an outlier in either the opening or closing operation of the platform doors, the outliers for each opening and closing operation will appear consecutively in either the opening or closing operation. By performing regression analysis on datasets obtained through repeated processes, it is possible to understand how the data changes over time. Furthermore, if the dataset is repeatedly collected over a long period, it is possible to identify long-term trends and patterns, which can be used to make future predictions.

[0024] Next, the information processing method according to Embodiment 1 will be described. Figure 4 is a flowchart showing the information processing method according to Embodiment 1. The extraction unit 11 extracts information on at least some of the values ​​from the results of regression analysis of the dataset used for training (step ST1). The output unit 12 outputs information about the process from which the dataset was acquired, specifically the process from which data corresponding to some of the values ​​extracted by the extraction unit 11 was acquired (step ST2). For example, the output unit 12 outputs display information for displaying the process information to the display device 2. After step ST2, the display device 2 displays the process information based on the display information from the output unit 12. The information processing device 1 can provide process information, which is information that improves the interpretability of the regression-analyzed dataset, by performing the series of processes shown in Figure 4.

[0025] (Variation 1) The extraction unit 11 extracts information on some of the values ​​of the explanatory variables, determined based on the predicted values ​​of the target variable. For example, the extraction unit 11 uses a tree-based model such as a decision tree or a random forest to calculate the importance of each explanatory variable and determines which explanatory variables are important for prediction based on the calculated importance. Features with high importance are likely to provide important information for predicting the target variable. Furthermore, for example, the extraction unit 11 may quantify the extent to which each explanatory variable contributes to the prediction of the dependent variable as a SHAP (SHapley Additive exPlanations) value, and extract information on some values ​​that have a small contribution to the prediction by the regression model based on the SHAP value. This makes it possible for the information processing device 1 to extract information on some values ​​determined based on the predicted value of the dependent variable from the regression analysis results.

[0026] Alternatively, when a dataset is input, the extraction unit 11 may use a machine learning model that outputs information on some values ​​determined based on the predicted values ​​of the target variable for each explanatory variable to extract information on some values ​​from the results of regression analysis of the dataset. The machine learning model may be stored in a memory unit (not shown in Figure 1) provided by the information processing device 1, or it may be stored in an external memory device that is separate from the information processing device 1 and capable of communication connectivity. In this case, the extraction unit 11 establishes a communication connection with the external storage device by controlling the communication unit. The machine learning model stored in the external storage device receives the dataset transmitted from the extraction unit 11 by the communication unit as input, and the information output from the machine learning model is received by the communication unit and acquired by the extraction unit 11 or the output unit 12.

[0027] (Modification 2) Furthermore, the extraction unit 11 may extract information on some of the values ​​of the explanatory variables that fall outside the confidence interval containing the predicted value of the dependent variable. For example, the extraction unit 11 checks whether the value of each point in the regression analysis results shown in Figure 2 is within the confidence interval C1. If the value of point F is outside the confidence interval C1, the extraction unit 11 detects point F as an outlier. An outlier may be determined by how far it is from the upper or lower limit of the confidence interval. The extraction unit 11 extracts information on the value of point F that is outside the confidence interval C1. This makes it possible for the information processing device 1 to extract information on some of the values ​​from the regression analysis results that fall outside the confidence interval C1 containing the predicted value E of the dependent variable.

[0028] (Variation 3) The extraction unit 11 may extract information on some values ​​determined by the user from the results of regression analysis of the dataset. For example, the user sets the values ​​of a specific explanatory variable in the extraction unit 11 using the input device 3. The extraction unit 11 extracts information on some values ​​corresponding to the explanatory variable values ​​set by the user from the regression analysis results. This makes it possible for the information processing device 1 to extract information on some values ​​determined by the user from the regression analysis results.

[0029] (Modification 4) Furthermore, the extraction unit 11 may extract information on some of the values ​​for the explanatory variables that fall outside the user-determined confidence interval, which includes the predicted value of the dependent variable. For example, the output unit 12 displays the regression analysis results shown in Figure 2 on the display device 2. The user, referring to the information displayed on the display device 2, sets a new confidence interval using the input device 3. The extraction unit 11 extracts information on some of the values ​​that fall outside the confidence interval determined by the user. In this way, the information processing device 1 can extract information on some of the values ​​that fall outside the confidence interval determined by the user from the regression analysis results.

[0030] (Variation 5) Furthermore, the output unit 12 may output to the display device 2 information of the process for which data corresponding to some of the values ​​extracted by the extraction unit 11 was acquired. Figure 5 shows an overview of the outlier extraction process. For example, when the extraction unit 11 obtains the regression analysis results shown in Figure 2, it extracts information on the value of point F that falls outside the confidence interval C1 in the regression model. The output unit 12 generates display information for displaying Figure (1), which plots the information on the value of point F extracted by the extraction unit 11 on a space axis consisting of kilometers over time and the point values ​​from the regression analysis of vibration data, as shown in Figure 5, and outputs the generated display information to the display device 2. Based on this display information, the display device 2 displays Figure (1), as shown in Figure 3A, in which outlier points obtained each time the railway vehicle completes a loop around the track are plotted. The outliers are the aforementioned partial values ​​extracted from the regression analysis results by the extraction unit 11. In addition, among the datasets subjected to regression analysis, the data corresponding to the above outliers is the data corresponding to the information on the aforementioned partial values. By referring to Figure (1), it is possible to recognize outliers that appear in multiple loops.

[0031] The output unit 12 identifies the kilometer markers where outliers appear consecutively in the relationship shown in Figure (1), and extracts the interval represented by the identified kilometer markers. The information of this interval is the process information for which data corresponding to the information of some of the values ​​extracted by the extraction unit 11 was acquired. The output unit 12 generates display information for displaying Figure (2), which shows the interval in which outlier points appear consecutively, and Figure (2) is displayed on the display device 2 based on this display information. By referring to Figure (2), it is possible to recognize the interval in which outlier points appear consecutively.

[0032] (Experimental variation 6) Furthermore, the output unit 12 may output to the display device 2 information showing the relationship between the process information where data corresponding to the information of some values ​​extracted by the extraction unit 11 was acquired and the information of some values. As described above, in Figure 5, the outliers are the some values ​​extracted from the regression analysis results by the extraction unit 11, and the data corresponding to the outliers in the dataset subjected to regression analysis is the data corresponding to the information of some values. The output unit 12 identifies a section B on the route map A on which the railway vehicle runs, which corresponds to a section in which outlier points appear consecutively, generates display information for displaying Figure (3) which distinguishes the identified section B from other sections of the route, and outputs the generated display information to the display device 2. Based on this display information, the display device 2 displays Figure (3) which links the section in which outlier points appear consecutively with section B on the route map A. Here, the information showing the section in which outlier points appear consecutively is the process information, and the information showing section B on the route map A is the information of some values. By referring to Figure (3), it is possible to recognize which part of the route map has outlier points appearing consecutively. For example, by referring to Figure (3), the cause of the outlier, such as the tracks being winding in section B, can be inferred, and the data corresponding to the outliers obtained in section B may be processed. Alternatively, by instructing the information processing device 1 using the input device 3, the data corresponding to the outliers obtained in section B may be removed from the regression analysis. The output unit 12 may display any of Figures (1), (2), and (3) shown in Modification 5 and Modification 6 on the display device 2, or it may display any combination of these.

[0033] Figure 6 shows the regression analysis results after processing outliers. In Figure 6, the horizontal axis represents the railway vehicle operation data, which is the explanatory variable, and the vertical axis represents the vibration data, which is the dependent variable. Confidence interval C2 is the confidence interval set by the regression analysis results after removing the outlier point F shown in Figure 2. Point D is a point obtained from the regression analysis of normal data that falls within confidence interval C2. Predicted value E is the predicted value of the dependent variable obtained from the regression analysis. Point F is a point obtained from the regression analysis of normal data that falls outside confidence interval C2. Comparing Figure 2 and Figure 6, it can be seen that by removing the data corresponding to the outlier obtained in part B, the number of points that fall outside confidence interval C2 is significantly reduced.

[0034] (Interval extraction) Next, we will explain in detail the process of extracting intervals in which outlier points appear consecutively. For example, the extraction unit 11 sets consecutive intervals based on the magnitude of the intervals between each value in the process from the values ​​obtained by regression analysis of the dataset, and extracts information on the values ​​included in the set intervals. This makes it possible to extract intervals in which each value appears consecutively. Figure 7A shows outlier points plotted on the kilometer axis. Figure 7B shows intervals extracted using a coarse distance criterion that include the outlier points plotted on the kilometer axis. Figure 7C shows intervals extracted using a fine distance criterion that include the outlier points plotted on the kilometer axis.

[0035] In the example shown in Figure 7A, the extraction unit 11 uses the distance between the furthest outlier point and the next furthest outlier point on the kilometer axis as distance reference m1, and the distance between the furthest outlier point and the next furthest outlier point on the kilometer axis as distance reference m2. The distance reference m1 or distance reference m2 may be set by the user using the input device 3, or it may be pre-set in the extraction unit 11. When extracting intervals using distance criterion m1, as shown in Figure 7B, the extraction unit 11 uses two adjacent points separated by a distance of m1 as a reference point in areas where multiple outlier points appear on the kilometer axis. If the outlier data adjacent to these points falls within distance criterion m1, the unit extracts the range of kilometers including these points as a continuous interval G1 where the outlier points are continuous. On the other hand, if the points are outside distance criterion m1, the unit determines that they are discontinuous. Similarly, when extracting intervals using distance criterion m2, as shown in Figure 7C, the extraction unit 11, in areas where multiple outlier points appear on the kilometer axis, uses two adjacent points separated by a distance of m2 as a reference. If the outlier points adjacent to these points fall within the distance criterion m2, the unit extracts the range of kilometers containing these outlier points as a continuous interval where the outlier points are consecutive. On the other hand, if they are outside the distance criterion m2, the unit determines that they are discontinuous. As a result, in the example in Figure 7C, continuous intervals G2 and G3 are extracted.

[0036] Next, we will describe the hardware configuration that realizes the functions of the information processing device 1. Figure 8 is a block diagram showing the hardware configuration that realizes the functions of the information processing device 1. The functions of the extraction unit 11 and output unit 12 of the information processing device 1 are realized by a processing circuit. That is, the information processing device 1 includes a processing circuit for executing the processes from step ST1 to step ST2 shown in Figure 4. The processing circuit may be a CPU (Central Processing Unit) that executes a program stored in memory.

[0037] The extraction unit 11, for example, acquires information indicating the results of regression analysis of a dataset used for learning, which is received from an external device by the communication unit of the information processing device 1 via the input interface 100. Furthermore, if the information processing device 1's storage unit stores information indicating the results of regression analysis of the dataset used for learning, the extraction unit 11 reads and acquires the information from the storage unit via the input interface 100. In this case, the information processing device 1 does not necessarily need to have a communication unit.

[0038] The output unit 12 outputs to the display device 2, for example, via the output interface 101, information on the process from which data corresponding to some of the values ​​extracted by the extraction unit 11 was acquired. The output unit 12 may also transmit the process results to an external device by controlling the communication unit via the output interface 101.

[0039] The functions of the extraction unit 11 and output unit 12 of the information processing device 1 are realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in memory 103.

[0040] The processor 102 realizes the functions of the extraction unit 11 and output unit 12 of the information processing device 1 by reading and executing a program stored in the memory 103. For example, the information processing device 1 includes a memory 103 for storing a program that, when executed by the processor 102, will result in the execution of the processes from step ST1 to step ST2 shown in Figure 4. These programs cause the computer to execute the procedures or methods of processing performed by the extraction unit 11 and output unit 12. The memory 103 may be a computer-readable storage medium that stores a program for causing the computer to function as the extraction unit 11 and output unit 12.

[0041] Memory 103 includes, for example, non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), and EEPROM (Electrically-EPROM) (registered trademark), as well as magnetic disks, flexible disks, optical disks, compact disks, minidiscs, and DVDs.

[0042] Some of the functions of the extraction unit 11 and output unit 12 of the information processing device 1 may be implemented by dedicated hardware, while other parts may be implemented by software or firmware. For example, the function of the extraction unit 11 may be realized by a processing circuit, which is dedicated hardware, and the function of the output unit 12 may be realized by the processor 102 reading and executing a program stored in the memory 103. In this way, the processing circuit can realize the above functions by hardware, software, firmware, or a combination thereof.

[0043] As described above, the information processing device 1 according to Embodiment 1 includes an extraction unit 11 that extracts information on at least some values ​​from the results of regression analysis of a dataset used for learning, and an output unit 12 that outputs information on the process in which the data corresponding to the information on some values ​​extracted by the extraction unit 11 was acquired. By referring to the process information, it is possible to understand the process in which the data corresponding to the information on some values ​​extracted from the data of the regression analysis results was acquired. Based on the understood process, for example, even if there is variation in the data corresponding to some values, it can be distinguished from other data. In this way, the information processing device 1 is capable of providing information that improves the interpretability of the regression-analyzed dataset.

[0044] In the information processing device 1 according to Embodiment 1, the extraction unit 11 extracts information on a portion of the values ​​of the explanatory variables that are determined based on the predicted values ​​of the dependent variable. As a result, the information processing device 1 can extract information on a portion of the values ​​determined based on the predicted values ​​of the dependent variable from the regression analysis results.

[0045] In the information processing device 1 according to Embodiment 1, the extraction unit 11 extracts information on some of the values ​​of the explanatory variables that fall outside the confidence interval that includes the predicted value of the dependent variable. As a result, the information processing device 1 can extract information on some of the values ​​from the regression analysis results that fall outside the confidence interval C1 that includes the predicted value E of the dependent variable.

[0046] In the information processing device 1 according to Embodiment 1, the extraction unit 11 extracts information on some values ​​determined by the user from the results of regression analysis of the dataset. This makes it possible for the information processing device 1 to extract information on some values ​​determined by the user from the regression analysis results.

[0047] In the information processing device 1 according to Embodiment 1, the extraction unit 11 extracts information on a subset of values ​​for the explanatory variables that fall outside the user-determined confidence interval, which includes the predicted value of the dependent variable. This allows the information processing device 1 to extract information on a subset of values ​​from the regression analysis results that fall outside the user-determined confidence interval.

[0048] In the information processing device 1 according to Embodiment 1, the dataset is acquired by repeatedly performing the process. By performing regression analysis on the dataset acquired by repeatedly performing the process, it is possible to understand how the data changes over time. Furthermore, if the dataset is acquired repeatedly over a long period of time, it is possible to understand long-term trends and patterns and use them to make future predictions.

[0049] In the information processing device 1 according to Embodiment 1, the output unit 12 outputs information of the process in which data corresponding to some of the values ​​extracted by the extraction unit 11 was acquired to the display device 2. By referring to this display information, the process in which data corresponding to some of the values ​​was acquired can be recognized.

[0050] In the information processing device 1 according to Embodiment 1, the output unit 12 outputs to the display device 2 information showing the relationship between the process information in which data corresponding to some of the extracted values ​​was obtained and the information of some of the extracted values. By referring to this display information, it is possible to recognize the information showing the relationship between the process information in which data corresponding to some of the extracted values ​​was obtained and the information of some of the extracted values.

[0051] In the information processing device 1 according to Embodiment 1, the extraction unit 11 sets consecutive intervals based on the magnitude of the intervals between each value in the process from the values ​​obtained by regression analysis of the dataset, and extracts information on the values ​​included in the set intervals. As a result, the information processing device 1 can extract intervals in which each value appears consecutively.

[0052] The information processing method according to Embodiment 1 includes the steps of: (ST1) an extraction unit 11 extracting information on at least some values ​​from the results of regression analysis of a dataset used for learning; and (ST2) an output unit 12 outputting information on the process in which the dataset was acquired, specifically the process in which the data corresponding to the values ​​extracted by the extraction unit 11 was acquired. As a result, the information processing method according to Embodiment 1 can provide information that improves the interpretability of the regression-analyzed dataset.

[0053] A computer executing the program according to Embodiment 1 performs the following steps: the extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for learning (ST1), and the output unit 12 outputs information on the process in which the data corresponding to the values ​​extracted by the extraction unit 11 was acquired (ST2). In this way, a computer executing the program according to Embodiment 1 functions as an information processing device 1 and can provide information that improves the interpretability of the regression-analyzed dataset.

[0054] Embodiment 2. In Embodiment 1, the extraction unit outputted information about the process in which data corresponding to some of the extracted values ​​was acquired. However, in Embodiment 2, it is possible to choose whether or not to use the extracted information for learning.

[0055] Figure 9 is a block diagram showing an example configuration of the information processing device 1A according to Embodiment 2. In Figure 9, the information processing device 1A extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1A acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1A may acquire regression analysis results stored in its own memory and generate process information based on the acquired information.

[0056] The learning device 4 uses the dataset output from the output unit 12A as training data to perform learning. For example, the learning device 4 is a regression analysis device that performs regression analysis using a regression model. In Figure 9, the learning device 4 is shown as an external device of the information processing device 1A, but the information processing device 1A may also be equipped with a learning unit that has the functions of the learning device 4.

[0057] The information processing device 1A comprises an extraction unit 11A, an output unit 12A, and a selection unit 13. For example, the information processing device 1A is implemented by a computer. The computer's memory stores programs that constitute information processing applications for realizing the functions of the extraction unit 11A, the output unit 12A, and the selection unit 13. The computer's processor executes the information processing applications read from the memory, thereby realizing the functions of the extraction unit 11A, the output unit 12A, and the selection unit 13.

[0058] (Selection section) The selection unit 13 accepts a selection of whether or not to use the information of some of the values ​​extracted by the extraction unit 11A for training. For example, suppose a user referring to Figure (3) shown in Figure 5 infers the cause of the outlier, such as the tracks being winding in section B, and determines that the regression model will overfit if the data corresponding to the outliers obtained in section B is used for training. In this case, the user uses the input device 3 to input setting information that the data corresponding to the outliers obtained in section B extracted by the extraction unit 11A will not be used for training. The selection unit 13 identifies the data acquired in section B based on the setting information input using the input device 3, and associates unavailable information with the identified data to instruct that it not be used for learning. The output unit 12A receives regression analysis results from the extraction unit 11A, which include information on some values ​​associated with unavailable information. Based on the unavailable information, it generates a dataset excluding data corresponding to outliers and outputs the generated dataset to the learning device 4 as training data. This allows the learning device 4 to perform training using training data with less variability.

[0059] Furthermore, if the user determines that using data corresponding to outliers obtained in part B for training will not result in overfitting, they input setting information using the input device 3 to instruct that the data corresponding to outliers obtained in part B extracted by the extraction unit 11A may be used for training. The selection unit 13 identifies the data acquired in section B based on the setting information input using the input device 3, and associates availability information with the identified data to indicate that it may be used for learning. The output unit 12A receives regression analysis results from the extraction unit 11A, which include information on some values ​​associated with available information. Based on the available information, it generates a dataset containing data corresponding to outliers and outputs the generated dataset to the learning device 4 as training data. This prevents unnecessary reduction of training data.

[0060] As described above, the information processing device 1A according to Embodiment 2 includes a selection unit 13 that accepts a selection of whether or not to use the information of some of the values ​​extracted by the extraction unit 11 for learning. By including the selection unit 13, the information processing device 1A can select whether or not to use the information of some of the values ​​extracted from the regression analysis results for learning.

[0061] Embodiment 3. In Embodiment 2, the user was given the option to choose whether or not to use information about some of the values ​​extracted from the regression analysis results for learning. However, in Embodiment 3, if the information about the process in which the data corresponding to the information about some of the values ​​was acquired conforms to the set conditions, it is automatically determined that the information about some of the values ​​will not be used for learning.

[0062] Figure 10 is a block diagram showing an example configuration of the information processing device 1B according to Embodiment 3. In Figure 10, the information processing device 1B extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1B acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1B may acquire regression analysis results stored in its own memory and generate process information based on the acquired information.

[0063] The information processing device 1B comprises an extraction unit 11, an output unit 12A, and a determination unit 14. For example, the information processing device 1B is implemented by a computer. The computer's memory stores programs that constitute information processing applications for realizing the functions of the extraction unit 11, the output unit 12A, and the determination unit 14. The computer's processor executes the information processing applications read from the memory, thereby realizing the functions of the extraction unit 11, the output unit 12A, and the determination unit 14.

[0064] (Judgment Department) The determination unit 14 determines whether the process information from which data corresponding to some of the values ​​extracted by the extraction unit 11 was obtained conforms to the set conditions. For example, suppose the determination unit 14 is set with a predetermined number of curves in the track within a certain length range, and route information indicating the shape of the track, and the output unit 12A outputs to the determination unit 14 information indicating part B in Figure (3) shown in Figure 5, for example, as process information indicating a section in which outlier data appears consecutively. In this case, the determination unit 14 determines whether the track in part B conforms to the set conditions based on the result of comparing the route information and the process information.

[0065] If the track in section B conforms to the set conditions, it is presumed that outlier points appear consecutively because the track in section B is winding. If the set conditions are met, the determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning. For example, the determination unit 14 outputs a dataset that does not include the data acquired in section B to the learning device 4. This allows the learning device 4 to perform learning using learning data with less variability.

[0066] If the track in section B does not conform to the set conditions, it is presumed that there are no factors causing the track to deviate in section B. If the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. For example, the determination unit 14 outputs a dataset containing the data acquired in section B to the learning device 4. This prevents unnecessary reduction of the training data.

[0067] (Setting conditions) The setting conditions indicate that the process in which data corresponding to some of the values ​​extracted from the regression analysis results was obtained is in a state where data variability is likely to occur. For example, the setting conditions may include changes in the external environment, structural or material non-uniformity, mechanical wear or deterioration, differences in operators or operating methods, or changes in measurement conditions or environmental conditions, which are expected to cause data variability.

[0068] Next, the information processing method according to Embodiment 3 will be described. Figure 11 is a flowchart showing the information processing method according to Embodiment 3. The extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for learning (step ST1A). The output unit 12A outputs information to the determination unit 14 about the process from which the dataset was acquired that obtained data corresponding to the values ​​extracted by the extraction unit 11 (step ST2A). The determination unit 14 determines whether the process information output from the extraction unit 11 conforms to the set conditions (step ST3A). The determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning if the set conditions are met. However, if the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. The information processing device 1B can provide process information, which improves the interpretability of the regression-analyzed dataset, by performing the series of processes shown in Figure 11, and can accurately determine whether or not to use some of the values ​​extracted from the regression analysis results for learning.

[0069] As described above, the information processing device 1B according to Embodiment 3 includes a determination unit 14 that determines whether the information of the process in which data corresponding to the information of some values ​​extracted by the extraction unit 11 was acquired conforms to the set conditions. If it conforms to the set conditions, the values ​​extracted by the extraction unit 11 are not used for learning. In this way, the information processing device 1B can automatically decide whether or not to use the data corresponding to the information of some values ​​extracted from the regression analysis results for learning.

[0070] Embodiment 4. Embodiment 3 automatically determined that some of the data corresponding to some of the values ​​would not be used for learning if the information of the process in which data corresponding to some of the values ​​was acquired conformed to the set conditions. However, Embodiment 4 labels the data corresponding to some of the values ​​that are not used for learning or the data corresponding to some of the values ​​that are not used for learning.

[0071] Figure 12 is a block diagram showing an example configuration of the information processing device 1C according to Embodiment 4. In Figure 12, the information processing device 1C extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1C acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1C may acquire regression analysis results stored in its own memory unit and generate process information based on the acquired information.

[0072] The information processing device 1C comprises an extraction unit 11, an output unit 12A, a determination unit 14, and a labeling unit 15. For example, the information processing device 1C is implemented by a computer. The computer's memory stores programs that constitute information processing applications for implementing each of the functions of the extraction unit 11, the output unit 12A, the determination unit 14, and the labeling unit 15. The computer's processor executes the information processing applications read from the memory, thereby realizing each of the functions of the extraction unit 11, the output unit 12A, the determination unit 14, and the labeling unit 15.

[0073] (Labeling section) The labeling unit 15 labels data that corresponds to some of the values ​​extracted by the extraction unit 11 but not used for training, or data that corresponds to some of the values ​​extracted by the extraction unit 11 but not used for training. For example, the labeling unit 15 assigns label information to points F that fall outside the confidence interval C1 in the regression model shown in Figure 2, which were extracted by the extraction unit 11, indicating that they will not be used as training data. Alternatively, the labeling unit 15 may assign label information to the data before regression analysis that corresponds to points F that fall outside the confidence interval C1 in the regression model shown in Figure 2, indicating that they will not be used as training data.

[0074] Next, the information processing method according to Embodiment 4 will be described. Figure 13 is a flowchart showing the information processing method according to Embodiment 4. The extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for training (step ST1B). The determination unit 14 determines whether the process information output from the extraction unit 11 conforms to the set conditions (step ST2B). The determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning if the set conditions are met. However, if the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. The labeling unit 15 labels data that contains some values ​​not used for learning, or data that corresponds to some values ​​not used for learning, based on the determination result from the determination unit 14 (step ST3B). The output unit 12A identifies, based on the label information, some values ​​that are not used for learning, or data corresponding to some values ​​that are not used for learning, and outputs a dataset that does not include data corresponding to some values ​​that are not used for learning to the learning device 4 (step ST4B). In addition, similar to Embodiment 1, the output unit 12A may output display information to the display device 2 for displaying information about the process in which data corresponding to some values ​​extracted by the extraction unit 11 was acquired among the processes in which the dataset was acquired. The display device 2 displays information about the process in which data corresponding to some values ​​was acquired based on the display information. The information processing device 1C can provide process information, which improves the interpretability of the regression-analyzed dataset, by performing the series of processes shown in Figure 13. Furthermore, based on label information, the information processing device 1C can accurately distinguish data corresponding to some values ​​extracted from the regression analysis results that are not used for training.

[0075] (modified version) The values ​​extracted by the extraction unit 11 may be trained using a different learning method than the values ​​that were not extracted. For example, the values ​​that were not extracted from the regression analysis results by the extraction unit 11 may be treated as values ​​obtained by the regression analysis, and the values ​​extracted from the regression analysis results by the extraction unit 11 may be treated as values ​​obtained by the k-nearest neighbors method. Generally, regression analysis is considered effective for datasets where the degree of deviation from normal values ​​is not large and the distribution is cohesive. This is because regression analysis is based on the assumption that there is a linear relationship between the explanatory variable and the dependent variable. When the deviation from the normal range is not large and the data is cohesive, a linear relationship can often be observed, allowing the data to be appropriately modeled in regression analysis. The data corresponding to values ​​that were not extracted from the regression analysis results by the extraction unit 11 can be said to be data with few large outliers and a cohesive distribution. For example, if the learning unit 16 identifies a value that was not extracted from the regression analysis results by the extraction unit 11 based on the label information assigned by the labeling unit 15, it performs regression analysis on the dataset corresponding to the identified value.

[0076] On the other hand, the k-nearest neighbors method, when given a new data point, classifies that data point into the class that appears most frequently among the k nearest training data points. Because the k-nearest neighbors method is not complex, it is possible to model nonlinear relationships. The data corresponding to the values ​​extracted from the regression analysis results by the extraction unit 11 can be said to be data with many outliers and lacking coherence as a distribution. For example, if the learning unit 16 determines, based on the label information assigned by the labeling unit 15, that a value has been extracted from the regression analysis results by the extraction unit 11, it performs learning using the k-nearest neighbors method on the dataset corresponding to the determined value. This allows the information processing device 1C to adopt different learning algorithms depending on the data trends.

[0077] While regression analysis and the k-nearest neighbors method were shown, the learning method is not limited to these, as long as it is appropriate for the trends in the data.

[0078] As described above, the information processing device 1C according to Embodiment 4 includes a labeling unit 15 that labels data corresponding to some of the values ​​extracted by the extraction unit 11 that are not used for learning, or data corresponding to some of the values ​​extracted by the extraction unit 11 that are not used for learning. Based on the label information, the information processing device 1C can accurately distinguish data corresponding to some of the values ​​extracted from the regression analysis results that are not used for learning.

[0079] In the information processing device 1C according to Embodiment 4, the values ​​extracted by the extraction unit 11 are learned using a different learning method than the values ​​that were not extracted. This makes it possible for the information processing device 1C to adopt different learning algorithms depending on the trends of the data.

[0080] Embodiment 5. While embodiments 2 to 4 describe information processing devices connected to a learning device, embodiment 5 describes an information processing device equipped with a learning unit.

[0081] Figure 14 is a block diagram showing an example configuration of the information processing device 1D according to Embodiment 5. In Figure 14, the information processing device 1D extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1D acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1D may acquire regression analysis results stored in its own memory and generate process information based on the acquired information.

[0082] The information processing device 1D comprises an extraction unit 11, an output unit 12, a determination unit 14, and a learning unit 16. For example, the information processing device 1D is implemented by a computer. The computer's memory stores programs that constitute information processing applications for implementing each of the functions of the extraction unit 11, the output unit 12, the determination unit 14, and the learning unit 16. The computer's processor executes the information processing applications read from the memory, thereby realizing each of the functions of the extraction unit 11, the output unit 12, the determination unit 14, and the learning unit 16.

[0083] (Learning Department) The learning unit 16 performs learning based on a dataset that excludes values ​​extracted by the extraction unit 11, either based on settings from the user or the judgment results from the judgment unit 14. For example, the learning unit 16 identifies some values ​​that are not used for learning, or data corresponding to some values ​​that are not used for learning, based on the determination result of the determination unit 14, and uses the dataset excluding the identified data as training data to perform learning such as degradation detection. Furthermore, for example, if the information processing device 1D is connected to the input device 3 shown in Figure 1, the user can use the input device 3 to set values ​​from the values ​​extracted by the extraction unit 11 that will not be used for learning. For example, values ​​that will not be used for learning can be set by linking information using the input device 3. The learning unit 16 performs learning using the dataset excluding the data set by the user as the learning data.

[0084] Next, the information processing method according to Embodiment 5 will be described. Figure 15 is a flowchart showing the information processing method according to Embodiment 5. The extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for training (step ST1C). The determination unit 14 determines whether the process information output from the extraction unit 11 conforms to the set conditions (step ST2C). The determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning if the set conditions are met. However, if the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. The learning unit 16 performs learning based on a dataset that excludes the values ​​extracted by the extraction unit 11, for example, based on the judgment result of the judgment unit 14 (step ST3C). The output unit 12 outputs the learning results from the learning unit 16 (step ST4C). For example, when the output unit 12 obtains the regression analysis results from the learning unit 16, it generates display information for displaying the regression analysis results and outputs the generated display information to the display device 2. The display device 2 displays the regression analysis results from the learning unit 16 based on the display information. In addition, the output unit 12 may, as in Embodiment 1, display on the display device 2 information about the process from which data corresponding to some of the values ​​extracted by the extraction unit 11 was obtained among the processes from which the dataset was acquired. The information processing device 1D can provide process information, which is information that improves the interpretability of the regression-analyzed dataset, by performing a series of processes shown in Figure 15. Furthermore, the information processing device 1D can perform learning using training data with low variability.

[0085] (Variation 1) Figure 16 is a block diagram showing the configuration of an information processing device 1E, which is a modified example of the information processing device 1D. In Figure 16, the information processing device 1E extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1E acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1E may acquire regression analysis results stored in its own memory and generate process information based on the acquired information.

[0086] The information processing device 1E comprises an extraction unit 11, an output unit 12, a determination unit 14, a learning unit 16, and a degradation determination unit 17. For example, the information processing device 1E is implemented by a computer. The computer's memory stores programs that constitute information processing applications for implementing each of the functions of the extraction unit 11, output unit 12, determination unit 14, learning unit 16, and degradation determination unit 17. The computer's processor executes the information processing applications read from the memory, thereby realizing each of the functions of the extraction unit 11, output unit 12, determination unit 14, learning unit 16, and degradation determination unit 17.

[0087] (Deterioration judgment section) The degradation determination unit 17 performs degradation determination based on the trained model trained by the learning unit 16. For example, the trained model trained by the learning unit 16 is a machine learning model that outputs degradation determination results from regression analysis when a dataset is input. The degradation determination result is data that indicates whether the result of the regression analysis of the data is data that has been caused by the degradation of the data being analyzed (hereinafter referred to as degraded data). For example, degraded data is data that deviates significantly from the value corresponding to normal data by exceeding an acceptable threshold. The degradation determination unit 17 performs degradation determination by inputting a dataset to the trained model that has been trained by the learning unit 16.

[0088] Furthermore, the degradation determination unit 17 may identify processes where only outliers corresponding to variable degradation data appear by comparing processes where outliers corresponding to degradation data appear, even though the data is normal, with processes where outliers corresponding to degradation data appear. This allows information on processes where only degradation data falls outside the confidence interval to be obtained, reducing not only over-detection but also missed degradation.

[0089] Next, a modified example 1 of the information processing method according to Embodiment 5 will be described. Figure 17 is a flowchart showing a modified example 1 of the information processing method according to Embodiment 5. The extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for training (step ST1D). The determination unit 14 determines whether the process information output from the extraction unit 11 conforms to the set conditions (step ST2D). The determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning if the set conditions are met. However, if the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. The learning unit 16 performs learning based on a dataset that excludes the values ​​extracted by the extraction unit 11, for example, based on the judgment result of the judgment unit 14 (step ST3D). The degradation determination unit 17 performs degradation determination based on the trained model that has been trained by the learning unit 16 (step ST4D). The output unit 12 outputs the degradation determination result from the degradation determination unit 17 (step ST5D). For example, when the output unit 12 obtains the degradation determination result from the degradation determination unit 17, it generates display information for displaying the degradation determination result and outputs the generated display information to the display device 2. The display device 2 displays the degradation determination result from the degradation determination unit 17 based on the display information. In addition, the output unit 12 may, as in Embodiment 1, display on the display device 2 information about the process from which data corresponding to some of the values ​​extracted by the extraction unit 11 was obtained among the processes from which the dataset was acquired. The information processing device 1E can provide process information, which is information that improves the interpretability of the regression-analyzed dataset, by performing a series of processes shown in Figure 17. Furthermore, the information processing device 1E can perform degradation detection based on a trained model obtained through training using training data with low variability.

[0090] (Modification 2) The output unit 12 may output to the display device 2 information on the process corresponding to the value determined to be degraded by the degradation determination unit 17 in the dataset input to the trained model. Figure 18 shows an overview of the process for presenting data for degradation candidates. In Figure 18, the model is a learning model that simulates the time evolution of vibrations occurring in a railway vehicle running on the route shown in route map A, and is generated by learning using dataset (1). After this model is generated, for example, after about 10 years, the route environment will have changed, and it will be necessary to rebuild the trained model. Dataset (2) is assumed to be data from about 10 years after the model was generated using dataset (1).

[0091] By performing training using dataset (2), the trained model shown in the center of Figure 18 is generated. Region B1 in route map A is a region where outlier points, which are normal data but outside the confidence interval, appear consecutively. Dataset (1) is obtained by removing the variable data acquired from railway vehicles traveling in region B1. In other words, the information for region B1 is information about the process in which data corresponding to the outliers extracted from the regression analysis data by the extraction unit 11 was acquired.

[0092] The degradation determination unit 17 performs degradation determination based on the trained model described above. For example, the degradation determination unit 17 takes the dataset (2) as input and outputs the result of degradation determination using the trained model. As a result of this degradation determination, regions B1 and B2 are determined to be degradation candidates, where the route is predicted to be degraded. Dataset (2) includes data obtained from region B1 and data obtained from region B2.

[0093] The degradation determination unit 17 outputs information indicating regions B1 and B2 to the output unit 12 as process information corresponding to the values ​​determined to be degraded in the dataset (2). The output unit 12 generates display information showing areas B1 and B2 on route map A, and outputs the generated display information to the display device 2. Based on this display information, the display device 2 displays route map A as shown in Figure 18. By referring to this route map A, it is possible to identify processes in which data with variations appears.

[0094] Furthermore, the degradation determination unit 17 can distinguish between a process in which data corresponding to an outlier point that is normal data but outside the confidence interval is acquired, and a process in which degraded data is acquired, based on the degree of deviation from normal data. In Figure 18, the data acquired from region B1 is data corresponding to an outlier and not degraded data, while the data acquired from region B2 is degraded data.

[0095] Therefore, the degradation determination unit 17 may associate the data obtained from region B1 with label information indicating that it corresponds to an outlier, since the data obtained from region B1 is not degradation data but data corresponding to an outlier. Based on the label information, the output unit 12 generates display information that shows areas B1 and B2 on route map A, along with information indicating that the data acquired from area B1 is not degraded, and outputs the generated display information to the display device 2. Based on this display information, the display device 2 displays route map A as shown in Figure 18. By referring to this route map A, it can be understood that area B1 is a process in which data corresponding to an outlier appears. The learning unit 16 may also train the trained model using the dataset (2) obtained by excluding the data acquired from region B1 based on the label information.

[0096] As described above, the information processing device 1D according to Embodiment 5 includes a learning unit 16 that performs learning based on a dataset excluding values ​​extracted by the extraction unit 11, based on settings from the user or determination results from the determination unit 14. This enables the information processing device 1D to perform learning using learning data with less variability.

[0097] The information processing device 1E according to Embodiment 5 includes a degradation determination unit 17 that performs degradation determination based on a trained model that has been trained by the learning unit 16. Degradation determination can be performed based on a trained model obtained by training using training data with little variation.

[0098] In the information processing device 1E according to Embodiment 5, the output unit 12 outputs information on the process corresponding to the value determined to be degraded by the degradation determination unit 17 in the dataset input to the trained model to the display device 2. By referring to the displayed information, it is possible to identify the process in which data with variability appears.

[0099] Embodiment 6. Embodiment 4 describes an information processing device that labels data corresponding to some values ​​that are not used for training or some values ​​that are not used for training. Embodiment 6 describes an information processing device that removes labeled data from a dataset input to a trained model.

[0100] Figure 19 is a block diagram showing the configuration of the information processing device 1F according to Embodiment 6. In Figure 19, the information processing device 1F extracts information on at least some values ​​from the regression analysis results and outputs information on the process from which data corresponding to the extracted information on some values ​​was acquired among the processes from which the dataset was acquired. For example, the information processing device 1F acquires regression analysis results from an external device via a network and generates process information based on the acquired information. The external device may be a regression analysis device that performs regression analysis on a pre-set dataset. Alternatively, the information processing device 1F may acquire regression analysis results stored in its own memory and generate process information based on the acquired information.

[0101] The information processing device 1F comprises an extraction unit 11, an output unit 12B, a determination unit 14, a labeling unit 15, a learning unit 16, and a removal unit 18. For example, the information processing device 1F is implemented by a computer. The computer's memory stores programs that constitute information processing applications for implementing each of the functions of the extraction unit 11, output unit 12B, determination unit 14, labeling unit 15, learning unit 16, and removal unit 18. The computer's processor executes the information processing applications read from memory, thereby realizing each of the functions of the extraction unit 11, output unit 12B, determination unit 14, labeling unit 15, learning unit 16, and removal unit 18.

[0102] (Removal part) The removal unit 18 removes labeled data from the dataset that is input to the trained model. For example, the labeling unit 15 labels data that has been extracted by the extraction unit 11 but is not used for training, or data that corresponds to values ​​extracted by the extraction unit 11 but is not used for training. The removal unit 18 removes data to which labeling information has been assigned by the labeling unit 15 from the data that constitutes the training dataset.

[0103] Next, the information processing method according to Embodiment 6 will be described. Figure 20 is a flowchart showing the information processing method according to Embodiment 6. The extraction unit 11 extracts information on at least some values ​​from the results of regression analysis of the dataset used for training (step ST1E). The determination unit 14 determines whether the process information output from the extraction unit 11 conforms to the set conditions (step ST2E). The determination unit 14 does not use the values ​​extracted by the extraction unit 11 for learning if the set conditions are met. However, if the set conditions are not met, the determination unit 14 may use the values ​​extracted by the extraction unit 11 for learning. The labeling unit 15 labels data that contains some values ​​not used for learning, or data that corresponds to some values ​​not used for learning, based on the determination result from the determination unit 14 (step ST3E). The removal unit 18 removes the labeled data from the dataset that is input to the trained model (step ST4E). The learning unit 16 performs training using the dataset from which some data has been removed by the removal unit 18 (step ST5E). The output unit 12 outputs the learning results from the learning unit 16 (step ST6E). For example, when the output unit 12 obtains the regression analysis results from the learning unit 16, it generates display information for displaying the regression analysis results and outputs the generated display information to the display device 2. The display device 2 displays the regression analysis results from the learning unit 16 based on the display information. In addition, the output unit 12 may, as in Embodiment 1, display on the display device 2 information about the process from which data corresponding to some of the values ​​extracted by the extraction unit 11 was obtained among the processes from which the dataset was acquired. The information processing device 1F can provide process information, which is information that improves the interpretability of the regression-analyzed dataset, by performing the series of processes shown in Figure 20. Furthermore, the information processing device 1F can perform learning using training data with less variability.

[0104] As described above, the information processing device 1F according to Embodiment 6 includes a labeling unit 15 that labels values ​​extracted by the extraction unit 11 that are not used for learning, or data corresponding to values ​​extracted by the extraction unit 11 that are not used for learning, and a removal unit 18 that removes labeled data from the dataset input to the trained model. As a result, the information processing device 1F can perform learning using training data with less variability.

[0105] Furthermore, it is possible to combine each embodiment, modify any component of each embodiment, or omit any component in each embodiment. [Industrial applicability]

[0106] The information processing device relating to this disclosure can be used, for example, in various systems or devices that perform repetitive operations. [Explanation of symbols]

[0107] 1,1A,1B,1C,1D,1E,1F Information processing device, 2 Display device, 3 Input device, 4 Learning device, 11,11A Extraction unit, 12,12A,12B Output unit, 13 Selection unit, 14 Judgment unit, 15 Labeling unit, 16 Learning unit, 17 Degradation judgment unit, 18 Removal unit, 100 Input interface, 101 Output interface, 102 Processor, 103 Memory.

Claims

1. An extraction unit that extracts information on at least some values ​​from the results of regression analysis on the dataset used for training, An output unit outputs information about the process in which data corresponding to the information of some of the values ​​extracted by the extraction unit was obtained from among the processes in which the aforementioned dataset was acquired, An information processing device equipped with the following features.

2. The extraction unit extracts information on some of the values ​​of the explanatory variables, which are determined based on the predicted values ​​of the dependent variable. The information processing apparatus according to feature 1.

3. The extraction unit extracts information on some of the values ​​of the explanatory variable that fall outside the confidence interval that includes the predicted value of the dependent variable. The information processing apparatus according to feature 2.

4. The extraction unit extracts information on some of the values ​​determined by the user from the results of regression analysis of the dataset. The information processing apparatus according to feature 1.

5. The extraction unit extracts information on some of the values ​​that fall outside the user-determined confidence interval for the explanatory variable, which includes the predicted value of the dependent variable. The information processing apparatus according to feature 4.

6. The aforementioned dataset is obtained by repeatedly performing the process. The information processing apparatus according to any one of claims 1 to 5.

7. The output unit outputs to the display device information of the process in which data corresponding to the information of the partial values ​​extracted by the extraction unit was obtained. The information processing apparatus according to any one of claims 1 to 5.

8. The output unit outputs to a display device information showing the relationship between the process information from which data corresponding to the information of the partial value extracted by the extraction unit was obtained, and the information of the partial value. The information processing apparatus according to feature 7.

9. The system includes a selection unit that accepts the choice of whether or not to use the information of the partial values ​​extracted by the extraction unit for learning. The information processing apparatus according to any one of claims 1 to 5.

10. The system includes a determination unit that determines whether the information of the process from which data corresponding to the information of the partial values ​​extracted by the extraction unit was obtained conforms to the set conditions, If the above setting conditions are followed, the values ​​extracted by the extraction unit will not be used for learning. The information processing apparatus according to any one of claims 1 to 5.

11. The system includes a labeling unit that labels data corresponding to the information of some of the values ​​extracted by the extraction unit but not used for learning, or the information of some of the values ​​extracted by the extraction unit but not used for learning. The information processing apparatus according to feature 10.

12. The values ​​extracted by the extraction unit are learned using a different learning method than the values ​​that were not extracted. The information processing apparatus according to feature 10.

13. The learning unit includes a learning unit that performs learning based on the dataset, excluding values ​​extracted by the extraction unit, based on settings from the user or the determination result from the determination unit. The information processing apparatus according to feature 10.

14. The system includes a degradation determination unit that performs degradation determination based on a trained model that has been trained by the aforementioned learning unit. The information processing apparatus according to feature 13.

15. The output unit outputs to the display device information the process corresponding to the value determined to be degraded by the degradation determination unit in the dataset input to the trained model. The information processing apparatus according to feature 14.

16. A labeling unit that labels data corresponding to values ​​extracted by the extraction unit but not used for learning, or values ​​extracted by the extraction unit but not used for learning, The trained model includes a removal unit that removes labeled data from the dataset input to the trained model. The information processing apparatus according to feature 14.

17. The extraction unit sets consecutive intervals based on the magnitude of the intervals between each value in the process from the values ​​obtained by regression analysis of the dataset, and extracts information on the values ​​included in the set intervals. The information processing apparatus according to any one of claims 1 to 5.

18. An information processing method performed by an information processing device, The extraction unit includes the step of extracting information on at least some values ​​from the results of regression analysis of the dataset used for training, The output unit outputs information about the process in which data corresponding to the values ​​extracted by the extraction unit was obtained, among the processes in which the dataset was acquired. An information processing method comprising the following:

19. On the computer, The extraction unit includes the step of extracting information on at least some values ​​from the results of regression analysis of the dataset used for training, The output unit outputs information about the process in which data corresponding to the values ​​extracted by the extraction unit was obtained, among the processes in which the dataset was acquired. A program to execute.

Images