JP7628926B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program for time-series data.

In information processing using machine learning and other methods, preprocessing to remove outliers from the large amount of training data used for learning is considered important. This is because if the training data contains outliers, the learning results may be unintended or the learning efficiency may decrease.

Methods for removing outliers differ depending on whether the training data is cross-sectional data that represents information at a single point in time, or time-series data along a time axis. For cross-sectional data, the Smirnoff-Grubbs test and Hotelling's theory are well known. These methods use the mean and variance of the data as parameters, making it difficult to apply them to time-series data, where the parameters of the data change dynamically from time to time.

An outlier in time series data can be (1) a time within a single time series sample that exhibits an abnormal value, or (2) an unusual sample within multiple time series samples at a single time.

Regarding the outliers in (1) above, Japanese Patent Application Laid-Open No. 2008-117381 (Patent Document 1) discloses a time series data analysis device that aims to realize a time series data analysis device capable of optimal modeling. This device detects abnormal values in a sample by taking the difference between the original value in the sample and the value smoothed in the time series.

For the outliers in (2) above, WO 2016/116961 (Patent Document 2) discloses an information processing device for more accurately determining the time when a time series signal begins to show an abnormality. This device detects an abnormality based on upper and lower limits of normal data that are set in advance.

JP 2008-117381 A International Publication No. 2016/116961

However, no matter which outlier removal method is used, it is not possible to evaluate the changes that a given piece of time series data exhibits along the time axis in comparison with other samples included in the training data. In addition, when the number of observation points in the time series data is small, it becomes difficult to smooth the time series data or express it using a statistical model such as an autoregressive model.

In this specification, time-series data with a small number of observation time points is referred to as longitudinal data. The present invention aims to realize information processing characterized by outlier removal from a large amount of longitudinal learning data used for learning such as machine learning, which confirms both the uniqueness of each time point compared to other samples and the uniqueness of intra-sample variation.

An information processing apparatus according to one aspect of the invention disclosed in the present application includes a processor that executes a program and a storage device that stores the program, the storage device stores a plurality of time series data, each of which includes data at two or more points in time, the processor calculates a rate of change with respect to time for each of the plurality of time series data, compares a distribution of the rate of change with a predetermined probability distribution, and determines, as peculiar data, the time series data for which a degree of deviation of the rate of change from the predetermined probability distribution satisfies a condition for peculiar data determination , thereby relatively evaluating the rate of change of the plurality of time series data, compares the distribution of the rate of change of the plurality of time series data excluding the time series data determined to be peculiar data with the predetermined probability distribution while changing the condition for peculiar data determination so that the number or a ratio of the time series data determined to be peculiar data satisfies the predetermined condition, and determines, as peculiar data, the time series data for which a degree of deviation of the rate of change from the predetermined probability distribution satisfies the condition for peculiar data determination .

According to a representative embodiment of the present invention, samples can be relatively evaluated based on the rate of change of each sample along the time axis from time-course learning data used in machine learning and the like, and unique samples can be extracted. Problems, configurations, and effects other than those described above will be made clear through the explanation of the following examples.

1 is a block diagram showing an example of a hardware configuration of an information processing device according to an embodiment of the present invention. 1 is a block diagram showing an example of a functional configuration of an information processing device according to an embodiment of the present invention. FIG. 4 is an explanatory diagram showing an example of the stored contents of notification information according to an embodiment of the present invention. FIG. 11 is an explanatory diagram showing an example of non-continuous value data according to an embodiment of the present invention. FIG. 1 is an explanatory diagram illustrating an example of continuous value data according to an embodiment of the present invention. FIG. 11 is an explanatory diagram illustrating an example of change rate data according to an embodiment of the present invention. FIG. 4 is an explanatory diagram illustrating an example of an input/output screen according to an embodiment of the present invention. FIG. 13 is an explanatory diagram illustrating an example of an execution result screen according to the embodiment of the present invention.

The information processing device according to the present invention will be described below with reference to the attached drawings. In this specification, an example of information processing aimed at removing anomalous data for predicting insurance payment risk in underwriting assessment of life insurance is shown. In the underwriting assessment, the future insurance payment risk is assessed based on information notified by the insurance applicant (hereinafter, notified information), and a decision is made as to whether to approve or reject the insurance application. The notified information includes the results of a health checkup, a medical interview, medical history, etc.

<Example of hardware configuration of information processing device>
FIG. 1 is a block diagram showing an example of a hardware configuration of an information processing apparatus according to an embodiment of the present invention.

The information processing device 100 has a processor 101, a storage device 102, an input device 103, an output device 104, and a communication interface (communication IF) 105. The processor 101, the storage device 102, the input device 103, the output device 104, and the communication IF 105 are connected by a bus 106. The processor 101 controls the information processing device 100. The storage device 102 is the working area of the processor 101. The storage device 102 is a non-temporary or temporary recording medium that stores various programs and data. Examples of the storage device 102 include a ROM (Read Only Memory), a RAM (Random Access Memory), a HDD (Hard Disk Drive), and a flash memory. The input device 103 inputs data. The input device 103 may be, for example, a keyboard, a mouse, a touch panel, a numeric keypad, or a scanner. The output device 104 outputs data. The output device 104 may be, for example, a display or a printer. The communication IF 105 connects to a network (not shown) and transmits and receives data.

<Example of functional configuration of information processing device 100>
FIG. 2 is a block diagram showing an example of a functional configuration of the information processing device 100 according to an embodiment of the present invention.

The information processing device 100 has an input unit 201, a data cleansing unit 202, a data classification unit 203, a time-dependent change rate calculation unit 204, a peculiar sample removal unit 205, and an output unit 206. Specifically, the data cleansing unit 202, the data classification unit 203, the time-dependent change rate calculation unit 204, and the peculiar sample removal unit 205 are realized, for example, by having the processor 101 execute a program stored in the storage device 102 shown in FIG. 1.

The information processing device 100 also stores the notification information 300, the non-continuous value data 400, the continuous value data 500, and the change rate data 600 in the storage device 102. The notification information 300 may be stored in advance in the information processing device 100, or may be acquired from another computer capable of communicating with the information processing device 100. First, the notification information 300 input to the input unit 201 will be described in detail.

Figure 3 is an explanatory diagram showing an example of the stored contents of notification information 300 in an embodiment of the present invention.

The notification information 300 is information necessary for the insurance contract notified by the contract applicant, and is the data to be analyzed. The notification information 300 has basic notification information 310, medical examination results 320, interview results 330, and medical history 340. The basic notification information 310 is basic information regarding the notification of the contract applicant. The basic notification information 310 includes a name ID 311, a date of birth 312, and an age 313.

Name ID 311 is identification information that uniquely identifies the contract applicant. Date of birth 312 is the date on which the contract applicant was born. The three entries of the contract applicant with name ID 311 "0001" in FIG. 3 show the analysis target data for the past three years for that contract applicant. Age 313 is the number of years that have passed since the contract applicant's date of birth 312. In the example described below, for the three entries of the contract applicant with name ID 311 "0001", age 313 is "47" for the first year in the chronological order, "48" for the second year in the chronological order, and "49" for the third year in the chronological order.

The medical checkup result 320 is the result of a medical checkup that the contract applicant underwent. The medical checkup result 320 includes weight 321, BMI (Body Mass Index) 322, systolic blood pressure 323, diastolic blood pressure 324, and fasting blood glucose 325. Weight 321 is the body weight of the contract applicant. BMI 322 is a body mass index that indicates the degree of obesity of a person, and is calculated by weight/(height ² ). The smaller the BMI 322 value, the thinner the person is, and the larger the BMI value, the fatter the person is.

Systolic blood pressure 323 is the pressure exerted on the aortic vascular wall by blood pushed out by cardiac contraction when blood is being pumped from the heart to the aorta. Diastolic blood pressure 324 is the pressure exerted on the aortic vascular wall when blood is returning to the heart, which is reduced as blood flows from the aorta into the heart due to cardiac expansion, reducing the amount of blood in the aorta. Fasting blood glucose 325 is the blood glucose level measured in a fasting state.

Medical interview result 330 is the result of the medical interview taken by the contract applicant. Medical interview result 330 includes smoking habits 331, drinking habits 332, and exercise habits 333. Smoking habits 331 are whether or not the contract applicant smokes, how often, and how much they smoke. Drinking habits 332 are whether or not the contract applicant drinks alcohol, how often, and how much they drink. Exercise habits 333 are whether or not the contract applicant exercises, how often, and how much they exercise.

Medical history 340 is a history of medical treatments the contract applicant has already received or been hospitalized. Medical history 340 includes hypertension medical examination history 341, hypertension hospitalization history 342, and diabetes medical examination history 343. Hypertension medical examination history 341 is a history of medical treatments the contract applicant has received for hypertension. Hypertension hospitalization history 342 is a history of hospitalizations the contract applicant has received for hypertension. Diabetes medical examination history 343 is a history of medical treatments the contract applicant has received for diabetes.

Referring again to FIG. 2. The data cleansing unit 202 extracts noise samples from the notification information 300 output from the input unit 201 and removes the noise. This makes it possible to improve the accuracy of removing anomalous data and the prediction accuracy of the subsequent insurance payment risk prediction. Types of noise samples include missing data and erroneous data input, and cleansing is achieved by deleting and correcting the respective samples.

The data classification unit 203 divides the cleansed notification information 300 output from the data cleansing unit 202 into continuous value data 500 and non-continuous value data 400 based on the data type. The continuous value data 500 is classified into weight 321, BMI 322, and the like. The non-continuous value data 400 is classified into binary data such as smoking habit 331, and categorical data such as drinking habit 332. This enables processing based on the data type, and achieves high accuracy in removing anomalous data.

Figures 4 and 5 are explanatory diagrams showing examples of non-continuous value data 400 and continuous value data 500, respectively, according to an embodiment of the present invention.

The non-continuous value data 400 is a subset of the notification information 300 and includes data other than the continuous value type data. For example, from the notification information 300 shown in FIG. 3, the non-continuous value data 400 including smoking habits 331, drinking habits 332, and exercise habits, which are category data indicating whether the habits correspond to "none," "once a week," or "twice a week," as well as hypertension consultation history 341, hypertension hospitalization history, and diabetes consultation history 343, which are binary data of "yes" or "no," are associated with the name ID 311 to generate the non-continuous value data 400.

The continuous value data 500 is a subset of the notification information 300 and includes continuous value type data such as weight 321, BMI 322, systolic blood pressure 323, diastolic blood pressure 324, and fasting blood glucose 325, which are associated with the name ID 311.

The continuous value data 500 is a collection of time series data to be evaluated by the information processing device 100 of this embodiment. For example, data for each item (e.g., weight, BMI, etc.) observed at multiple time points for each person constitutes one piece of time series data. However, although the number of observation time points for the time series data in this embodiment is not limited, it is expected to be smaller than that expected for general time series data, for example, 2 or more, or at most about 10. Such time series data may be described as longitudinal data.

Referring again to FIG. 2, the time-dependent change rate calculation unit 204 calculates the rate of change over time for each item of the continuous value data 500. Here, the number of time points included in the continuous value data 500 is n, and each time is t (1, 2, ..., T). Furthermore, if the continuous value data 500 includes M types of continuous value data, for example, an arbitrary continuous value variable at time t for a sample with a name ID 311 of 0001 can be expressed as V(0001, t, m). Here, m is a value between 1 and M, and identifies an item of continuous value data such as weight or BMI.

The time-dependent change rate calculation unit 204 calculates the rate of change U from the previous time from time t=2 onwards, from V(0001,2,m) to V(0001,T,m). The rate of change U at time t can be expressed by the following formula.

U(0001,t,m)=100×(V(0001,t,m) - V(0001,t-1,m))/V(0001,t-1,m)

The calculated rate of change is stored in the rate of change data 600. In this embodiment, the change over time of the continuous value data 500 is expressed as a rate of change, but the change over time can be quantified by expressing the change in time in any manner, such as the amount of change from the previous time, or the rate of change from two times before.

Figure 6 is an explanatory diagram showing an example of change rate data 600 relating to an embodiment of the present invention.

The change rate data 600 includes a name ID 601 that identifies each contract applicant, and the change rate of the continuous value data of each contract applicant at each time. As an example, FIG. 6 shows the change rate 602 at time 2, the change rate 603 at time T-1, and the change rate 604 at time T for the samples of continuous value data of item m with name IDs 601 of 0001 and 0002.

The peculiar sample exclusion unit 205 retrieves the rate of change U calculated by the time-dependent change rate calculation unit 204 from the rate of change data 600, determines whether each sample is a sample that exhibits a peculiar change compared to other samples, and excludes a sample from the notification information 300 if the change over time of the sample is peculiar.

If it is assumed that the change rates U of all samples follow a normal distribution, the peculiar sample exclusion unit 205 compares the distribution of the change rates U with the normal distribution, extracts samples that deviate from the normal distribution, and displays them through the output unit 206. To compare the distribution of the change rates U with the normal distribution, for example, a QQ (Quantile-Quantile) plot is used. A QQ plot is known as a method for visually checking whether the change rate U follows a normal distribution by plotting theoretical quantiles of the normal distribution corresponding to the quantiles of the change rate U.

In a QQ plot, if the line smoothly connecting the plotted quantiles forms a straight line, it can be determined that the rate of change U follows a perfect normal distribution. Conversely, when the plotted quantiles are approximated by a straight line, points that deviate from the straight line can be determined to be outliers from the normal distribution. The peculiar sample exclusion unit 205 calculates the confidence interval of the QQ plot based on multiple confidence coefficients, and excludes samples that deviate from any confidence interval as peculiar samples. This makes it possible to improve the accuracy of future insurance payment risk predictions.

One example of an automatic method of extracting peculiar samples in the peculiar sample exclusion unit 205 is to repeat the process until the extraction rate or number of samples to be excluded reaches a pre-specified value. Specifically, the peculiar sample exclusion unit 205 calculates a confidence interval with a sufficiently large reliability coefficient, such as 0.999, and extracts samples that deviate from the reliability interval. If the extraction rate or number of samples extracted in this way reaches a pre-specified value, the process ends; if not, the reliability coefficient is reduced and samples that deviate from the reliability interval are extracted again. This process is repeated, and the process ends when the extraction rate or number of samples reaches a pre-specified value.

Alternatively, the peculiar sample exclusion unit 205 may repeat the process of excluding the sample with the maximum deviation from the normal distribution, comparing the distribution of the rate of change U of the remaining samples with the normal distribution that it follows, and eliminating the sample with the maximum deviation from the normal distribution. This repetition may be performed until the extraction rate or number of samples to be excluded reaches a pre-specified value, or until the deviation from the normal distribution falls within a predetermined range (for example, until there are no samples that deviate from the confidence interval of a predetermined confidence coefficient).

The peculiar sample exclusion unit 205 can display the QQ plot on the output unit 206, and the user of the information processing device 100 can visually extract and exclude peculiar samples. In this embodiment, it is assumed that the rate of change U follows a normal distribution, but this is just one example, and a method of drawing a QQ plot by assuming any probability distribution other than the normal distribution may be adopted. If it is difficult to specify an arbitrary probability distribution in advance, a method of preparing a large number of probability distribution candidates and searching for a probability distribution that has the smallest removal sample extraction rate or number of samples extracted by repeated processing (i.e., the smallest degree of deviation of the rate of change U) may be adopted.

In addition, instead of using a QQ plot, a method can be considered in which peculiar samples are extracted and removed by visually checking the histogram. The set of change rates U processed by the peculiar sample removal unit 205 can be divided and processed according to the time changes at the same time for M types of items, or it can be processed together according to the time changes over the same time span, ignoring the time.

<Screen example>
FIG. 7 is an explanatory diagram showing an example of an input/output screen according to an embodiment of the present invention.

The input/output screen 700 is displayed on a display, which is an example of the output device 104 of the information processing device 100, or on the display of another computer (not shown) that can communicate with the information processing device 100.

The input/output screen 700 includes a notification information load button 701, a processing item selection pull-down 705, a reliability coefficient selection bar 708, a processing execution button 710, and an execution result display area 800. The notification information 300 may be loaded when the user operates the notification information load button 701. Alternatively, when the user operates the notification information input button 702, the information processing device 100 may display a notification information input screen (not shown), and the user may input the notification information using the input device 103.

Figure 8 is an explanatory diagram showing an example of an execution result screen related to an embodiment of the present invention.

The execution result display area 800 includes a plot screen 801 that displays the QQ plot drawn by the peculiar sample exclusion unit 205, and a peculiar sample ID screen 802 that displays the name ID of the extracted peculiar sample.

The plot screen 801 shows the quantiles of the rate of change U on the vertical axis and the theoretical quantiles of the normal distribution on the horizontal axis, with the points where the rate of change U is plotted being represented by black circles. If the rate of change U exactly follows the normal distribution, all the black circles are plotted on the thick solid line. The thin solid lines represent the confidence intervals calculated based on multiple confidence coefficients. The peculiar sample ID screen 802 is a screen that lists samples that deviate from the confidence intervals calculated based on the confidence coefficient specified by the confidence coefficient selection bar 708.

In addition, while the above-mentioned embodiment has been described using the example of insurance payment risk prediction in life insurance underwriting assessment, the present invention can also be applied to, for example, corporate financial analysis. In this case, the data described in the securities report or index data calculated from said data is used for evaluation instead of the notification information 300. In general, the present invention can be applied to extract peculiar samples (outliers) from a collection of multiple time series data when the number of observation points for each piece of time series data is small.

Furthermore, the functions of the information processing device 100 in the above-described embodiment may be provided via an API (Application Programming Interface). For example, when the information processing device 100 receives notification information 300 via a network connected to the communication IF 105, it may store it in the storage device 102, execute the process shown in FIG. 2, and output the notification information from which the anomalous sample has been removed and data required to display the information shown in FIG. 7 and FIG. 8 via the network.

The system of the present invention may also be configured as follows:

(1) An information processing device (e.g., information processing device 100) having a processor (e.g., processor 101) that executes a program and a storage device (e.g., storage device 102) that stores the program, the storage device stores multiple time series data (e.g., continuous value data 500) each including data from two or more points in time, and the processor calculates the rate of change over time for each of the multiple time series data (e.g., processing by the time-varying change rate calculation unit 204) and relatively evaluates the rates of change of the multiple time series data (e.g., processing by the unusual sample exclusion unit 205).

This makes it possible to evaluate samples relative to each other based on the rate of change of each sample over time, using longitudinal learning data used in machine learning and other applications.

(2) In the above (1), the processor compares the distribution of the change rates with a predetermined probability distribution, and determines that time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies the condition for determining peculiar data is peculiar data, thereby relatively evaluating the change rates of multiple time series data (e.g., processing by the peculiar sample exclusion unit 205).

This makes it possible to extract peculiar samples from longitudinal learning data used in machine learning and other applications by evaluating samples relative to each other based on the rate of change of each sample over time.

(3) In the above (2), the processor determines that the time series data having the highest deviation in rate of change from a predetermined probability distribution is anomalous data.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(4) In the above (3), the processor compares the distribution of the rates of change of multiple time series data excluding the time series data determined to be peculiar data with a predetermined probability distribution, and determines the time series data whose rate of change deviates most from the predetermined probability distribution as peculiar data. This process is repeated until the number or proportion of time series data determined to be peculiar data satisfies a predetermined condition or the degree of deviation of the rate of change from the predetermined probability distribution falls within a predetermined range.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(5) In the above (2), the processor repeatedly executes a process of comparing the distribution of the change rates of the multiple time series data excluding the time series data determined to be peculiar data with a predetermined probability distribution while changing the conditions for determining peculiar data so that the number or proportion of time series data determined to be peculiar data satisfies the predetermined condition, and determining as peculiar data the time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies the condition for determining peculiar data.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(6) In the above (2), the processor generates multiple probability distributions by assuming that the distributions of the change rates are based on different probability distribution models, compares the distribution of the change rates with the multiple probability distributions, and determines the probability distribution with the smallest degree of deviation of the change rates as the predetermined probability distribution to be compared for determining peculiar data.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(7) In (2) above, the predetermined probability distribution is a distribution based on a predetermined probability distribution model that the distribution of the rate of change is assumed to follow (e.g., normal distribution and other distributions),

The processor determines that time series data whose rate of change falls outside a confidence interval corresponding to a predetermined confidence coefficient of a predetermined probability distribution is anomalous data.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(8) In the above (7), the processor repeatedly performs a process of comparing the distribution of the change rates of the multiple time series data excluding the time series data determined to be peculiar data with a predetermined probability distribution while changing the reliability coefficient so that the number or proportion of time series data determined to be peculiar data satisfies a predetermined condition, and determining that the time series data whose change rate falls outside the reliability interval is peculiar data.

This makes it possible to properly remove anomalous samples from longitudinal training data used in machine learning and other applications.

(9) In the above (2), the information processing device further has an output device, and the output device displays a quantile-quantile plot based on the distribution of the change rate and a predetermined probability distribution.

This makes it easier to visually check for the presence of anomalous data.

The present invention is not limited to the above-described embodiments, but includes various modified examples and equivalent configurations within the spirit of the appended claims. For example, the above-described embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to having all of the configurations described. Furthermore, a portion of the configuration of one embodiment may be replaced with the configuration of another embodiment. Furthermore, the configuration of another embodiment may be added to the configuration of one embodiment. Furthermore, other configurations may be added, deleted, or replaced with part of the configuration of each embodiment.

Furthermore, each of the configurations, functions, processing units, processing means, etc. described above may be realized in part or in whole in hardware, for example by designing them as integrated circuits, or may be realized in software by having the processor 101 interpret and execute a program that realizes each function.

Information such as programs, tables, and files that realize each function can be stored in a storage device such as a memory, hard disk, or SSD (Solid State Drive), or in a computer-readable non-transitory data recording medium such as an IC (Integrated Circuit) card, SD card, or DVD (Digital Versatile Disc).

In addition, the control lines and information lines shown are those considered necessary for explanation, and do not necessarily represent all control lines and information lines necessary for implementation. In reality, it is safe to assume that almost all components are interconnected.

100 Information processing device 101 Processor 102 Storage device 202 Data cleansing unit 203 Data classification unit 204 Time-dependent change rate calculation unit 205 Unique sample removal unit 300 Notification information 400 Non-continuous value data 500 Continuous value data 600 Change rate data

Claims (6)

An information processing device,
A processor for executing a program and a storage device for storing the program,
the storage device holds a plurality of time series data each including data at two or more points in time;
The processor,
Calculating a rate of change over time for each of the plurality of time series data;
comparing the distribution of rates of change to a predetermined probability distribution;
Relatively evaluating the change rates of the plurality of time series data by determining, as peculiar data, that the degree of deviation of the change rate from the predetermined probability distribution satisfies a condition for peculiar data determination ;
an information processing apparatus which repeatedly executes a process of comparing a distribution of the change rates of the plurality of time series data excluding the time series data determined to be peculiar data with the predetermined probability distribution while changing a condition for the peculiar data determination so that the number or a ratio of the time series data determined to be peculiar data satisfies a predetermined condition, and determining, as peculiar data, the time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies the condition for the peculiar data determination . 2. The information processing device according to claim 1,
The processor,
generating a plurality of probability distributions by assuming that the distributions of the rates of change are based on different probability distribution models;
comparing the distribution of the rate of change to the plurality of probability distributions;
determining, as the predetermined probability distribution to be compared for determining whether data is anomalous, the probability distribution having the smallest degree of deviation from the rate of change. An information processing device,
A processor for executing a program and a storage device for storing the program,
the storage device holds a plurality of time series data each including data at two or more points in time;
The processor,
Calculating a rate of change over time for each of the plurality of time series data;
comparing the distribution of rates of change to a predetermined probability distribution;
Relatively evaluating the change rates of the plurality of time series data by determining, as peculiar data, that the degree of deviation of the change rate from the predetermined probability distribution satisfies a condition for peculiar data determination;
the predetermined probability distribution is a distribution based on a predetermined probability distribution model which the distribution of the change rate is assumed to follow;
The processor determines, as anomalous data, the time series data in which the rate of change falls outside a confidence interval corresponding to a predetermined confidence coefficient of the predetermined probability distribution;
an information processing apparatus which repeatedly executes a process of comparing a distribution of the change rates of the plurality of time series data excluding the time series data determined to be peculiar data with the predetermined probability distribution while changing the reliability coefficient so that the number or ratio of the time series data determined to be peculiar data satisfies a predetermined condition, and determining that the time series data whose change rate falls outside the reliability interval is peculiar data. 2. The information processing device according to claim 1,
further comprising an output device;
The information processing apparatus according to claim 1, wherein the output device displays a quantile-quantile plot based on the distribution of the change rates and the predetermined probability distribution. An information processing method executed by an information processing device,
The information processing device includes a processor that executes a program and a storage device that stores the program;
the storage device holds a plurality of time series data each including data at two or more points in time;
The information processing method includes:
A step of the processor calculating a rate of change with respect to time for each of the plurality of time series data;
the processor comparing the distribution of the rate of change to a predetermined probability distribution;
a step of relatively evaluating the change rates of the plurality of time series data by determining, by the processor, that the time series data is peculiar data when the degree of deviation of the change rate from the predetermined probability distribution satisfies a condition for peculiar data determination, and
an information processing method comprising: a step of relatively evaluating the change rates of the plurality of time series data; a step of repeatedly executing a process of comparing a distribution of the change rates of the plurality of time series data excluding the time series data determined to be peculiar data with the predetermined probability distribution while changing a condition for the peculiar data determination so that a number or a ratio of the time series data determined to be peculiar data satisfies a predetermined condition; and determining, as peculiar data, the time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies the condition for the peculiar data determination. An information processing program for controlling an information processing device,
the information processing device includes a processor that executes the information processing program and a storage device that stores the information processing program;
the storage device holds a plurality of time series data each including data at two or more points in time;
The information processing program includes:
calculating a rate of change over time for each of the plurality of time series data;
comparing said distribution of rates of change to a predetermined probability distribution;
a step of relatively evaluating the change rates of the plurality of time series data by determining, as peculiar data, the time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies a condition for peculiar data determination;
the step of relatively evaluating the change rates of the plurality of time series data, the information processing program repeatedly executes a process of comparing a distribution of the change rates of the plurality of time series data excluding the time series data determined to be peculiar data with the predetermined probability distribution while changing a condition for the peculiar data determination so that a number or a ratio of the time series data determined to be peculiar data satisfies a predetermined condition, and determining, as peculiar data, the time series data whose degree of deviation of the change rate from the predetermined probability distribution satisfies the condition for the peculiar data determination.

Images