Estimation device, estimation method, and program

Abstract

An estimation device and the like are provided that can output state transition probabilities that take into account the tendency of data change in data transitions between distributions. [Solution] An estimation device according to one aspect of the present disclosure includes a receiving means for receiving input of a data set for each stage, an estimation means for estimating a state transition probability based on a transition from a first distribution, which is the distribution of data in a first stage, to a second distribution, which is the distribution of data in a second stage, based on a cost required for the transition to the destination, which is determined for each data in the first distribution, and an output means for outputting the estimated state transition probability.

Description

[Technical Field]

[0001] The present disclosure relates to an estimation device, an estimation method, and a program. [Background technology]

[0002] BACKGROUND OF THE INVENTION Techniques exist for future analysis of information about people in fields such as healthcare.

[0003] Patent document 1 discloses that in clusters by age group, which are the result of clustering medical data, the transition probability between each cluster in the direction of age progression is calculated based on the similarity between the clusters, and by linking each cluster by age group with the transition probability between clusters, a predictive model is constructed that shows the transition of a subject's health condition by age group. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Application Laid-Open No. 2016-095684 Summary of the Invention [Problem to be solved by the invention]

[0005] When estimating a state transition, the ease of transition to the next transition may differ depending on the type of data being handled, the value of the data, etc. For example, test values ​​related to health status obtained through a health checkup or the like may tend to worsen with age more than they tend to improve. In this way, data may have a specific tendency to change.

[0006] One of the objectives of the present disclosure has been made in consideration of the above-mentioned problems, and is to provide an estimation device, etc. that is capable of outputting state transition probabilities that take into account the trend of data change when data transitions between distributions. [Means for solving the problem]

[0007] An estimation device according to one aspect of the present disclosure includes a receiving means for receiving input of a data set for each stage, an estimation means for estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for the transition to the destination, which is determined for each data in the first distribution, and an output means for outputting the estimated state transition probability.

[0008] The estimation method according to one aspect of the present disclosure includes receiving an input of a data set for each stage, The state transition probability based on a transition from a first distribution, which is the distribution of data in a first stage, to a second distribution, which is the distribution of data in a second stage, is estimated based on the cost required for the transition to the destination, which is determined for each data in the first distribution, and the estimated state transition probability is output.

[0009] A program according to one aspect of the present disclosure causes a computer to perform the following processes: accepting input of a data set for each stage; estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for the transition to the destination, which is determined for each data in the first distribution; and outputting the estimated state transition probability. [Effects of the Invention]

[0010] According to the present disclosure, it is possible to output a state transition probability that takes into account the tendency of data change when data transitions between distributions. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a first block diagram illustrating an example of a functional configuration of an estimation device according to the present disclosure. [Figure 2] 1 is a first flowchart illustrating an example of the operation of the estimation device of the present disclosure. [Figure 3]FIG. 2 is a second block diagram illustrating an example of the functional configuration of the estimation device of the present disclosure. [Figure 4] FIG. 10 is a diagram illustrating an example of a probability density distribution according to the present disclosure. [Figure 5] FIG. 1 is a diagram illustrating an image of solving a transition from one probability density distribution to another probability density distribution as an optimal transportation problem according to the present disclosure. [Figure 6] 10 is a second flowchart illustrating an example of the operation of the estimation device of the present disclosure. [Figure 7] FIG. 10 is a third block diagram illustrating an example of the functional configuration of the estimation device of the present disclosure. [Figure 8] FIG. 4 is a fourth block diagram illustrating an example of the functional configuration of the estimation device of the present disclosure. [Figure 9] 10 is a third flowchart illustrating an example of the operation of the estimation device of the present disclosure. [Figure 10] FIG. 5 is a fifth block diagram illustrating an example of the functional configuration of the estimation device of the present disclosure. [Figure 11] 10 is a fourth flowchart illustrating an example of the operation of the estimation device of the present disclosure. [Figure 12] 10 is a fifth flowchart illustrating an example of the operation of the estimation device of the present disclosure. [Figure 13] FIG. 2 is a block diagram illustrating an example of a hardware configuration of a computer device that realizes the estimation device of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0012] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0013] First Embodiment An outline of the estimation device of the first embodiment will be described.

[0014] The estimation device of the present disclosure uses accumulated data to calculate state transition probabilities between data. In the present disclosure, an example of target data is health data, which is data related to a person's health. The health data may be, for example, values ​​of test items in a health checkup or information related to a person's exercise habits. However, the health data is not limited to these examples.

[0015] The health data may be accumulated in advance, for example. In this case, the accumulated health data does not have to be longitudinal data resulting from continuous observation of a specific individual over a predetermined period of time. The accumulated health data may be data for each of multiple individuals at a predetermined point in time. For example, the results of health checkups for 10,000 individuals in a year t may be accumulated as health data. Note that, although the present disclosure mainly describes an example in which the estimation device estimates data transitions based on health data, the target data is not limited to this example.

[0016] 1 is a first block diagram showing an example of the functional configuration of the estimating device 100. As shown in FIG. 1, the estimating device 100 includes a receiving unit 110, an estimating unit 120, and an output unit .

[0017] The receiving unit 110 receives input of a dataset. The dataset may be health data of multiple people. For example, the dataset may be the results of health checkups conducted on multiple people over a one-year period. In this way, the dataset may be data related to the results of health checkups at a specific time point for each of multiple people. The dataset may be stored in advance in a storage device (not shown). The storage device may be a device included in the estimation device 100, or may be an external device communicatively connected to the estimation device 100.

[0018] Datasets are classified into stages. A stage may be information indicating a stratum when a dataset is stratified. In other words, a stage can also be information indicating a condition when population data is classified into subsets based on a predetermined condition. For example, if a dataset is health data, a dataset for each stage may be health data for each age group. More specifically, if the health data is data indicating the blood glucose levels of multiple people, a dataset for each stage may include data indicating the blood glucose levels of people in their teens, data indicating the blood glucose levels of people in their twenties, ..., and data indicating the blood glucose levels of people in their eighties. In other words, in this example, the dataset includes data indicating the blood glucose levels for each age group of ten years. In this way, the stages may be ordered. For example, the stage following the stage for people in their teens is the stage for people in their twenties. Note that the age groups may be any age group.

[0019] In this way, the receiving unit 110 receives input of a data set for each stage. The receiving unit 110 is an example of a receiving means.

[0020] The estimation unit 120 estimates state transition probabilities based on data transitions between predetermined stages. Specifically, the estimation unit 120 estimates state transition probabilities based on a transition from the data distribution in a first stage to the data distribution in a second stage. Here, the data distribution in the first stage is referred to as a first distribution. Also, the data distribution in the second stage is referred to as a second distribution.

[0021] At this time, the estimation unit 120 estimates the state transition probability based on the cost required for the transition to the transition destination, which is determined for each data. For example, the estimation unit 120 estimates which data in the first distribution will transition to in the second distribution. At this time, the cost required for the data transition is defined. The cost is determined depending on the transition destination. For example, the estimation unit 120 estimates the state transition probability by obtaining a pair of data in the first distribution and data in the second distribution that minimizes the total cost required for transitioning each data in the first distribution to the second distribution. At this time, the estimation unit 120 may estimate the state transition probability using an optimal transportation algorithm.

[0022] The cost is determined for each data item. Therefore, for example, the cost may be different for each data item even if the transition direction and distance are the same. Furthermore, for a given data item, the cost may be different depending on the transition direction even if the transition distance is the same. In other words, the cost may be asymmetric with respect to the increase or decrease in the data value. This makes it possible to determine a cost that takes into account cases where the ease of transition to the transition destination differs depending on the type or value of the data.

[0023] Assume that the health data is data indicating the blood glucose levels of multiple individuals. Assume also that the first stage is the age group in their 40s, and the second stage is the age group in their 50s. That is, the second stage is a stage subsequent to the first stage. In this case, the estimation unit 120 estimates, for example, a transition from the blood glucose level distribution of individuals in their 40s to the blood glucose level distribution of individuals in their 50s. Assume here that the data indicates that blood glucose levels tend to worsen and not improve with age. In this case, costs are set for blood glucose level data in the distribution for individuals in their 40s that tend to worsen and not improve. In other words, for blood glucose level data, a low cost is set for a worsening transition, and a high cost is set for an improvement transition.

[0024] For example, the estimation unit 120 uses the cost information to estimate a plausible transition from a probability distribution indicating the probability that there are people in their 40s who have each blood glucose value to a probability distribution indicating the probability that there are people in their 50s who have each blood glucose value. In this estimation, a mapping from the probability distribution for people in their 40s to the probability distribution for people in their 50s is estimated. Estimating this mapping is synonymous with estimating a state transition probability for a transition from the distribution of blood glucose levels of people in their 40s to the distribution of blood glucose levels of people in their 50s.

[0025] In this way, the estimation unit 120 estimates the state transition probability based on the transition from the first distribution, which is the distribution of data in the first stage, to the second distribution, which is the distribution of data in the second stage, based on the cost required for the transition to the transition destination, which is determined for each data in the first distribution. The estimation unit 120 is an example of an estimation means.

[0026] The output unit 130 outputs the estimated state transition probabilities. For example, the output unit 130 may output the state transition probabilities to an output device having a display or the like. Furthermore, for example, the output unit 130 may store the state transition probabilities in a storage device (not shown). The output unit 130 is an example of output means.

[0027] Next, an example of the operation of the estimation device 100 will be described with reference to Fig. 2. In this disclosure, each step in a flowchart will be represented by a number assigned to the step, such as "S1".

[0028] FIG. 2 is a flowchart illustrating an example of the operation of the estimation device 100.

[0029] The receiving unit 110 receives input of a data set for each stage (S1).

[0030] The estimation unit 120 estimates the state transition probability based on a transition from a first distribution, which is the distribution of data in the first stage, to a second distribution, which is the distribution of data in the second stage, based on the cost required for the transition to the destination, which is determined for each data in the first distribution (S2).

[0031] The output unit 130 outputs the estimated state transition probabilities (S3).

[0032] As described above, the estimation device 100 of the first embodiment receives an input of a data set for each stage. The estimation device 100 also estimates a state transition probability based on a transition from a first distribution, which is a distribution of data in the first stage, to a second distribution, which is a distribution of data in the second stage, based on a cost required for the transition to the transition destination, which is determined for each data in the first distribution. The estimation device 100 then outputs the estimated state transition probability.

[0033] For example, by setting a cost for each data item in the first distribution according to the ease of transition to the destination, the estimating device 100 can estimate a state transition probability based on a transition from the first distribution to the second distribution according to the trend of data change. That is, the estimating device 100 can output a state transition probability that takes into account the trend of data change in data transitions between distributions.

[0034] <Second embodiment> Next, an estimation device according to a second embodiment will be described. In the second embodiment, a further example of the estimation device described in the first embodiment will be described. In the second embodiment, an example in which the estimation device 100 estimates data transitions based on health data will also be mainly described, but the target data is not limited to this example. Note that some of the content overlapping with the first embodiment will be omitted.

[0035] [Details of the estimation device 100] 3 is a block diagram showing an example of the functional configuration of the estimation device 100. The estimation device 100 includes a receiving unit 110, an estimation unit 120, and an output unit 130. The estimation device 100 may also include an acquisition unit 140 and a classification unit 150. The estimation device 100 may also include a storage device 190. The storage device 190 may be a device included in the estimation device 100, or may be an external device communicatively connected to the estimation device 100.

[0036] The estimation device 100 is, for example, a device provided in a terminal device such as a personal computer. The terminal device is a device operated by a user. The estimation device 100 is not limited to this example, and may be a device realized in a server device communicably connected to the terminal device via a wired or wireless network. The estimation device 100 may perform various processes in response to instructions from the terminal device.

[0037] The estimation device 100 may also be communicatively connected to other devices via a wired or wireless network. For example, the estimation device 100 may be capable of communicating with an external server device that stores health data. The external server device may be, for example, a device managed by a hospital, a local government, or a company.

[0038] The receiving unit 110 receives input of a dataset for each stage. At this time, the dataset is stored in the storage device 190. For example, the receiving unit 110 may receive, as input of a dataset, reading a dataset stored in the storage device 190 in accordance with an instruction from a terminal device.

[0039] Health data is stored in storage device 190. Receiving section 110 may receive the health data as a data set. The health data is acquired by acquiring section 140.

[0040] The acquiring unit 140 acquires health data. Specifically, the acquiring unit 140 acquires the health data from an external server device that manages the health data. For example, assume that the external server device manages the health checkup results of 10,000 people in year t. The acquiring unit 140 acquires the health checkup results of 10,000 people in year t from the external server device as health data. The health data may be information corresponding to the test items in the health checkup. Furthermore, the health data is the results of health checkups taken by each of the 10,000 people at a certain point in time in year t. In this way, the data set may not be data showing changes over time in the same subject, but may be data measured at a single point in time for each of multiple subjects. The acquiring unit 140 stores the acquired health data in the storage device 190.

[0041] The method of acquiring health data is not limited to this example. For example, a recording medium storing health data may exist. In this case, the terminal device reads the health data from the recording medium. Then, acquisition unit 140 may acquire the health data read by the terminal device.

[0042] In this way, acquisition section 140 acquires health data, which is data relating to the health of each of a plurality of people at a predetermined point in time.

[0043] The health data is processed by, for example, classifying unit 150. The processed health data may then be stored in storage device 190. Classifying unit 150 processes the data acquired by acquiring unit 140 into a data set according to conditions. For example, classifying unit 150 classifies the health data by age group. For example, classifying unit 150 classifies the health data by age group of 10 years. However, this is not a limitation, and classifying unit 150 may classify the health data by any age interval. For example, classifying unit 150 may classify the health data by one year of age.

[0044] In this case, the classification unit 150 may extract specific data from the health data and classify the extracted data by age group. For example, assume that the health data includes information indicating height, weight, blood pressure, blood glucose level, HbA1c, and BMI (Body Math Index). In this case, the classification unit 150 may classify the data indicating blood glucose level and BMI from the health data by age group.

[0045] The classification conditions and the data to be extracted may be information according to instructions from a terminal device. That is, a user operating the terminal device inputs information indicating the classification conditions and the data to be extracted into the terminal device. The terminal device transmits the input information to the estimation device 100. The classification unit 150 processes the health data using the information indicating the classification conditions and the data to be extracted transmitted from the terminal device.

[0046] The classification unit 150 also generates a distribution related to the acquired data. Specifically, the classification unit 150 generates a probability density distribution for each condition based on the acquired data. For example, the classification unit 150 generates a distribution in which data indicating blood glucose levels and BMI are plotted for each age group. The distribution generated at this time is a two-dimensional distribution related to blood glucose levels and BMI. The classification unit 150 then generates a probability density distribution indicating the probability of existence of each value of blood glucose levels and BMI. The one-dimensional data value is expressed as x i Let the data values ​​of other dimensions be x j Then the probability density distribution is p([x i ,x j ]) In this way, the classifier 150 generates a probability density distribution for each age group regarding the acquired health data.

[0047] In this case, the classification unit 150 may classify data in each distribution into data groups. In this case, the probability density distribution is a distribution indicating the probability of existence of each data group. FIG. 4 is a diagram showing an example of a probability density distribution. The probability density distribution shown in FIG. 4 has blood glucose level and BMI as its axes. In addition, the example of FIG. 4 shows 64 cells. These cells are data groups into which data indicating each person's blood glucose level and BMI is classified. The probability of existence of each data group is shown. In this way, the classification unit 150 may classify data in each distribution of health data for each age group into data groups. Note that the classification unit 150 may generate a distribution with two or more dimensions. For example, the classification unit 150 may generate a probability density distribution with blood glucose level, BMI, and average number of steps per day as its axes. In addition, the classification unit 150 may generate a probability density distribution that is considered to change continuously by narrowing the width of the cells.

[0048] The receiving unit 110 receives the health data categorized by condition as a data set for each stage. For example, the receiving unit 110 may receive a probability density distribution for the health data for each age group as described above as a data set for each stage.

[0049] The estimation unit 120 estimates the state transition probability between stages. Specifically, the estimation unit 120 estimates the state transition probability based on the transition from the distribution of health data in a first age group to the distribution of health data in a second age group. Here, the second age group is an age group after the first age group. The distribution of health data in the first age group corresponds to the first distribution. The distribution of health data in the second age group corresponds to the second distribution.

[0050] The estimation unit 120 estimates the transition from the first distribution to the second distribution using an algorithm for the optimal transportation problem (hereinafter referred to as the optimal transportation algorithm). The optimal transportation algorithm is an algorithm for finding a transportation method that optimizes the cost required to transition from a given probability distribution to another probability distribution.

[0051] Specifically, for distributions μ and ν in the probability space X, the direct product X 2 The distribution π in is a coupling when the following equations 1 and 2 hold.

[0052]

number

[0053]

number

[0054] Let the total number of couplings be Π(μ,ν). Also, let c(x,y) be the cost function for transporting element x included in distribution μ to element y included in distribution ν. In this case, for example, in an equation such as the following equation 3, the coupling that minimizes the cost is called optimal transport.

[0055]

number

[0056] Alternatively, if the mapping from distribution μ to ν is T(x), then the direct transition can be found by finding T that minimizes the following equation (4). In this case, T is a one-to-one mapping. Also, for a subset U of μ, the mapping T(U) and U have the same volume.

[0057]

number

[0058] When optimal transport is performed for discrete data, it can be formulated as follows. Specifically, C ij Let be the cost matrix and the distribution be μ i , ν j In this case, under the condition shown in Equation 6, we can find P that minimizes Equation 5, which shows the total cost. ij Ask for.

[0059]

number

[0060]

number

[0061] In this way, the optimal transportation algorithm can calculate a set of data before transportation and data at the destination that optimizes the cost of transportation from the data distribution in the first stage to the data distribution in the second stage.

[0062] The estimation unit 120 estimates a state transition probability based on a transition from a first distribution to a second distribution. At this time, the estimation unit 120 solves the transition from the first distribution to the second distribution as an optimal transportation problem. Here, the first distribution and the second distribution are probability density distributions. That is, the estimation unit 120 solves the transition from a probability density distribution related to health data in a first age group to a probability density distribution related to health data in a second age group as an optimal transportation problem.

[0063] FIG. 5 is a diagram illustrating an example of solving the transition from one probability density distribution to another as an optimal transportation problem. FIG. 5 shows a probability density distribution for health data in a first age group and a probability density distribution for health data in a second age group. When the estimation unit 120 solves the transition from the first distribution to the second distribution as an optimal transportation problem, it is equivalent to estimating which of the cells in the probability density distribution for the second age group has a higher probability of transitioning from each cell in the probability density distribution for the first age group. In other words, the estimation unit 120 estimates state transition probabilities based on the transition from each data group in the first distribution to each data group in the second distribution.

[0064] For example, let μ be the probability density distribution for health data in a first age group, and v be the probability density distribution for health data in a second age group. In this case, the estimation unit 120 estimates the mapping T using, for example, Equation 4. For example, the estimation unit 120 models the mapping T as a function using a neural network or the like, such as a multi-layer fully connected layer. The estimation unit 120 also obtains the mapping T by performing optimization using machine learning so that Equation 3 becomes small. Furthermore, the estimation unit 120 uses this mapping T to generate multiple y's that transition from a given x. Then, the estimation unit 120 obtains the state transition probabilities from the generated y's. In this way, the estimation unit 120 estimates the state transition probabilities.

[0065] The output unit 130 outputs the estimated state transition probabilities. For example, the output unit 130 may output the estimated state transition probabilities to the storage device 190 and store the state transition probabilities in the storage device 190.

[0066] Next, a specific example of cost will be described. In the formulas such as Formula 3 to Formula 5, a formula related to cost, such as a cost function or a cost matrix, is defined. The cost may be defined according to the tendency of change in the data being handled. Hereinafter, the cost function and the cost matrix may be collectively referred to as the cost function.

[0067] [First example of cost] In the first example, we will explain an example in which the data being handled has a specific tendency to change as the data transitions from a first stage to a second stage. Specifically, assume that the health data is test values ​​from a health checkup or the like. For example, the health data is data related to blood glucose levels, HbA1c, and MMSE (Mini Mental State Examination). In this case, the health data tends to deteriorate more easily and improve less easily with age. Therefore, in the transition from the distribution of health data in a first age group (first distribution) to the distribution of health data in a second age group (second distribution), the cost required for the health data to deteriorate is set small, and the cost required for the health data to improve is set large.

[0068] In Equation 3, μ is the probability density distribution for health data in the first age group, and ν is the probability density distribution for health data in the second age group. Furthermore, let x be the health data in distribution μ, and y be the health data in distribution ν. For example, assume that the health data is data on HbA1c. In this case, an increase in the HbA1c value corresponds to a transition in which the health data worsens. Furthermore, a decrease in the HbA1c value corresponds to a transition in which the health data improves. In this case, the cost function can be expressed, for example, by Equation 7 or 8.

[0069]

number

[0070]

number

[0071] In Equation 7 and Equation 8, the value of function g(x) is reflected with respect to the squared error of the distance between elements x and y. Here, g(x) in Equation 7 and Equation 8 may be different functions. Any function can be set for g(x). In this example, a cost function is shown that reduces the cost when health data x in the first age group, which is the pre-transition stage, rises to health data y in the second age group, which is the post-transition stage. In other words, this cost function can be applied when an increase in health data corresponds to a transition in the direction of worsening health data. Note that when g(x) is set as a monotonically increasing function, it is expressed that the larger the value of health data x, the smaller the cost when health data x rises to transition to health data y. Similarly, by changing the role of g(x), it is possible to define a cost function that can be applied when an increase in health data corresponds to a transition in the direction of improving health data.

[0072] Furthermore, when optimal transportation is performed for discrete data, as in equation 6, the cost matrix can be expressed by the following equation 9.

[0073]

number

[0074] In this case, the first distribution and the second distribution may each represent a histogram showing the data among the seven types of value data.

[0075] As described above, in the first example, the cost function can be said to have asymmetry with respect to the increase and decrease of the value. Note that the above-described cost function is merely an example. In other words, the cost function is not limited to the above example.

[0076] In this way, the estimation unit 120 may estimate the state transition probability based on the transition from the first distribution to the second distribution based on a cost function such that the cost of the health data of the first distribution transitioning in a worsening direction is smaller than the cost of the health data of the first distribution transitioning in an improving direction when the health data of the first distribution transitions from the first distribution to the second distribution.

[0077] [Second example regarding costs] In the second example, in the transition of data between multidimensional distributions having at least two axes related to the data, a value corresponding to a certain axis has a specific tendency to change depending on the value corresponding to the other axis. For example, assume that the health data is HbA1c and the average number of steps per day (hereinafter simply referred to as "average number of steps"). That is, assume that the first distribution and the second distribution are probability density distributions with HbA1c and the average number of steps as their axes, respectively.

[0078] Here, it is assumed that the larger the value of the average number of steps, the less likely HbA1c will deteriorate and the more likely it will improve. In this case, a cost function is set in which the cost required for the transition from the first distribution to the second distribution varies depending on the value of the average number of steps.

[0079] In Equation 3, μ is the probability density distribution for health data in the first age group, and ν is the probability density distribution for health data in the second age group. Let x = (x1, x2) be the health data in distribution μ, and y = (y1, y2) be the health data in distribution ν. Here, the first axis is HbA1c, and the second axis is the average number of steps. That is, x1 and y1 are the HbA1c values, and x2 and y2 are the average number of steps. In this case, for example, if there is health data where the HbA1c value is a and the average number of steps is 5,000, and another health data where the HbA1c value is a and the average number of steps is 10,000, the cost required for the HbA1c value to increase (worsen) will be greater for the latter health data. In other words, the HbA1c value is less likely to increase for the latter health data. In this case, the cost function can be expressed by, for example, equation 10, equation 11, or equation 12.

[0080]

number

[0081]

number

[0082]

number

[0083] where d 2 is expressed as the following equation 13.

[0084]

number

[0085] In the formulas 10 to 12, the value of the function h is reflected with respect to the squared error regarding the distance between elements x and y. Here, the function h in each of the formulas 10 to 12 may be a different function. In this example, by setting h(x2) to a monotonically increasing function such that h(x2)>0, it is expressed that the larger the value of x2, the greater the cost required for transitioning in the direction of increasing the value of x1. In other words, it is expressed that in the health data of the first distribution, the larger the value of the average number of steps, the more difficult it is for the HbA1c value to transition in the direction of increasing (deteriorating). Similarly, by changing the role of the function h, it is possible to express that the smaller the value corresponding to a specific axis, the more difficult it is for the value corresponding to another axis to transition in the direction of increasing.

[0086] As described above, in the second example, the cost function can be said to be asymmetric with respect to a specific axis. By using a cost function that is asymmetric with respect to a specific axis, it becomes possible to estimate a state transition that depends on the value of the specific axis. Note that the above-mentioned cost function is merely an example. In other words, the cost function is not limited to the above-mentioned example. The first distribution and the second distribution may be three-dimensional or higher-dimensional distributions having three or more axes.

[0087] In this manner, the health data includes values ​​of multiple health-related items for each of multiple people. The first and second distributions are multidimensional distributions with each of the multiple health-related items as an axis. In this case, the estimation unit 120 may estimate the state transition probability based on the transition from the first distribution to the second distribution based on a cost function in which the cost of transitioning the health data from the first distribution to the second distribution varies depending on the magnitude of the value corresponding to a specific axis.

[0088] [Example of operation of the estimation device 100] Next, an example of the operation of the estimating device 100 will be described with reference to FIG.

[0089] FIG. 6 is a second flowchart illustrating an example of the operation of estimation device 100. Specifically, FIG. 6 is a flowchart illustrating an example of the operation of estimation device 100 when calculating a state transition probability between predetermined stages. In this example of operation, it is assumed that data indicating blood glucose levels and BMI are acquired as health data. In this case, it is assumed that blood glucose levels tend to worsen more easily and improve less easily as the BMI value increases. This example of operation also illustrates an example in which estimation device 100 calculates a state transition probability from the distribution of health data in the 40s age group to the distribution of health data in the 50s age group.

[0090] Acquiring unit 140 acquires health data. For example, acquiring unit 140 acquires health data from an external server device (S101). Acquiring unit 140 then stores the health data in storage device 190. Classifying unit 150 processes the health data (S102). For example, classifying unit 150 classifies the health data by age group. Then, classifying unit 150 generates a probability density distribution for each age group of 10 years for the health data. At this time, the probability density distribution is a distribution that indicates the probability of existence of each value of blood glucose level and BMI. In other words, the probability density distribution is a distribution with blood glucose level and BMI as its axes.

[0091] Receiving unit 110 receives input of health data for each age group (S103). For example, receiving unit 110 receives reading of health data stored in storage device 190 as input of health data.

[0092] The estimation unit 120 estimates the state transition probability based on a transition from the distribution of health data in a first age group (first distribution) to the distribution of health data in a second age group (second distribution) based on the cost function (S104). Here, the first age group is the 40s. The second age group is the 50s. That is, the estimation unit 120 estimates the state transition probability based on a transition from the probability density distribution for health data in the 40s to the probability density distribution for health data in the 50s. In this case, the cost function is a function that expresses the cost such that the higher the BMI value, the smaller the cost required for a transition in the direction of increasing (worsening) blood glucose level.

[0093] Then, the output unit 130 outputs the estimated state transition probability (S105). Specifically, the output unit 130 outputs the estimated state transition probability to the storage device 190 and stores the state transition probability in the storage device 190. Alternatively, the output unit 130 may output the estimated state transition probability to a terminal device having a display or the like.

[0094] Note that this operational example is merely an example. That is, the operation of the estimation device 100 of the present disclosure is not limited to this example. Furthermore, the estimation device 100 does not necessarily need to estimate state transition probabilities in consecutive stages. In this operational example, an example is shown in which the state transition probability is calculated from health data for people in their 40s to health data for people in their 50s based on health data categorized by age group in their teens. However, this example is not limiting, and for example, the estimation device 100 may calculate the state transition probability from health data for people in their 40s to health data for people in their 70s.

[0095] In this way, the estimation device 100 of the second embodiment accepts input of a data set for each stage. The estimation device 100 also estimates a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for the transition to the transition destination, which is determined for each data in the first distribution. The estimation device 100 then outputs the estimated state transition probability.

[0096] Specifically, estimation device 100 may receive health data for each age group, which is data related to the health of multiple people at a given time point, as a data set for each stage. In this case, the first distribution may be the distribution of the health data in the first age group, and the second distribution may be the distribution of the health data in a second age group that is an age group after the first age group.

[0097] For example, by setting a cost for each health data item in the first distribution according to the ease of transition to the next distribution, estimation device 100 can estimate a state transition probability based on a transition from the first distribution to the second distribution according to the trend of data change. In other words, estimation device 100 can output a state transition probability that takes into account the trend of data change when data transitions between distributions.

[0098] Furthermore, the estimation device 100 may estimate the state transition probability based on the transition from the first distribution to the second distribution based on a cost function such that the cost of the health data of the first distribution transitioning in a worsening direction is smaller than the cost of the health data of the first distribution transitioning in an improving direction. For example, suppose the health data has a tendency to deteriorate more easily and improve less easily with age. The estimation device 100 can estimate the state transition probability taking such a tendency of change in the health data into account.

[0099] Assume also that the health data includes values ​​of multiple health-related items for each of multiple individuals. Furthermore, assume that the first distribution and the second distribution are multidimensional distributions with each of the multiple health-related items as an axis. In such a case, the estimation device 100 may estimate a state transition probability based on a transition from the first distribution to the second distribution based on a cost function in which the cost of transitioning the health data from the first distribution to the second distribution varies depending on the magnitude of the value corresponding to a specific axis. This allows the estimation device 100 to estimate a state transition probability that takes into account changes in data that depend on the value corresponding to a specific axis.

[0100] [Variation 1] The above-described cost function may be designed in the estimating apparatus 100. FIG. 7 is a block diagram showing an example of a functional configuration of the estimating apparatus 100 according to Modification 1. In Modification 1, the estimating apparatus 100 may further include a design unit 160.

[0101] The design unit 160 designs the cost function in advance. For example, the design unit 160 designs the cost function using data that has been accumulated in advance.

[0102] Specifically, the design unit 160 designs the function g or the function h in Equations 7, 8, 10, 11, and 12. At this time, a function form is assumed in which coefficients are determined in advance as parameters. For example, the function g(x) can be expressed as Equation 14 below, where α, β, and γ are parameters.

[0103]

number

[0104] In addition, g(x) is expressed as a quadratic polynomial in Equation 14. However, g(x) is not limited to this example, and may be a linear, polynomial, exponential, or logarithmic function, or a combination thereof.

[0105] For example, the design unit 160 adjusts the parameters α, β, and γ in a cost function having a function g(x) of Equation 14 so that the total cost calculated by the cost function is minimized.

[0106] Furthermore, the function g(x) may be modeled using a neural network such as a multi-layer fully connected layer. In this case, the design unit 160 may obtain the modeled function g(x) by, for example, using the total cost as a loss function and performing optimization using machine learning. The function g may be optimized simultaneously with the mapping T. In this case, various machine learning techniques may be used.

[0107] Assume that health data is accumulated in advance. Furthermore, assume that the accumulated health data includes time-series data resulting from continuous observation of the same individual over a predetermined period of time. This time-series data is data that clearly shows the correspondence between age groups before and after a transition. Therefore, the design unit 160 may use the corresponding data, which is data that clearly shows the correspondence between the transitions, to design a cost function. For example, the design unit 160 uses changes in the distribution of the corresponding data in the age group before the transition and the distribution of the corresponding data in the age group after the transition, and adjusts the parameters α, β, and γ to match the changes.

[0108] Even in this case, the design unit 160 may model the function g(x) using a neural network or the like with multiple fully connected layers. The design unit 160 may use the difference between the corresponding data in the age group before the transition and the corresponding data in the mapped age group after the transition as a loss function and perform optimization using machine learning to obtain the modeled function g(x).

[0109] Furthermore, data other than the transition destination can be used as data that clearly shows the transition correspondence. For example, when learning about the transition from age 40 to age 50, the design unit 160 may use data for age 55. For example, even if there is no data with a long-term correspondence of 10 years or more, there may be data with a short-term correspondence of 5 years. In this case, the data used for 5 years in the future is not data that indicates the transition 10 years in the future. Therefore, the design unit 160 performs learning by virtually generating a transition 10 years in the future by doubling extrapolation to match the corresponding time.

[0110] Then, the estimation unit 120 estimates the state transition probability based on the transition from the first distribution to the second distribution, based on the designed cost function.

[0111] The design unit 160 can also design the function h in the same way.

[0112] In this way, the design unit 160 designs a cost function that calculates a cost determined for each data item based on a change in the data distribution between the data in the stage before the transition and the data in the stage after the transition. The design unit 160 is an example of a design means.

[0113] <Third embodiment> Next, an estimation device according to a third embodiment will be described. In the third embodiment, an example will be described in which the health state of a target person is predicted based on the calculated state transition probability. In the third embodiment, an example in which the estimation device estimates data transitions based on health data will also be mainly described, but the target data is not limited to this example. Note that some of the content overlapping with the first and second embodiments will not be described.

[0114] [Details of Estimation Device 101] The estimation device 101 is a device obtained by adding additional functional units to the estimation device 100. FIG. 8 is a block diagram showing an example of the functional configuration of the estimation device 101. The estimation device 101 includes a reception unit 110, an estimation unit 120, an output unit 130, an acquisition unit 140, and a classification unit 150. The estimation device 101 may also include a prediction unit 170. The estimation device 101 may also include a storage device 190.

[0115] Like the estimation device 100, the estimation device 101 may be a device provided in a terminal device, or may be a device implemented in a server device communicatively connected to the terminal device via a wired or wireless network.

[0116] Estimation device 101 predicts future health data for a target person using pre-calculated state transition probabilities. In this embodiment, the first age group is the 40s and the second age group is the 50s. The state transition probability from health data in the 40s to health data in the 50s is calculated.

[0117] The acquisition unit 140 acquires health data of the target person. For example, assume that the state transition probability is calculated from the probability density distribution related to blood glucose level and BMI. At this time, the acquisition unit 140 acquires health data indicating the blood glucose level and BMI of the target person. The acquired health data related to the target person is also referred to as target data.

[0118] Prediction unit 170 is an example of a prediction means. Prediction unit 170 predicts the transition of the target person's health data. In other words, prediction unit 170 predicts the future value of the target person's health data based on the target data. Specifically, prediction unit 170 predicts the value of the health data when the target person reaches an age corresponding to a second age group. In this case, the first age group is the age group corresponding to the target person's age in the target data.

[0119] For example, suppose the target person is 41 years old. In this case, the target person's age corresponds to the first age group. The prediction unit 170 identifies which data group the target data falls into among the probability density distributions related to health data for people in their 40s. Then, the prediction unit 170 predicts which data group the identified data group will transition to among the probability density distributions related to health data for people in their 50s, based on the state transition probability.

[0120] In this way, the prediction unit 170 predicts the transition of the target data, which is health data related to the target person. More specifically, the prediction unit 170 predicts, for example, a data group in the second distribution, which is a transition destination based on the state transition probability of the data group in the first distribution corresponding to the target data, as the health data when the target person reaches the age of the second age group.

[0121] [Example of operation of the estimation device 101] Next, an example of the operation of the estimation device 101 will be described with reference to FIG.

[0122] Fig. 9 is a third flowchart illustrating an example of the operation of the estimation device 101. Specifically, Fig. 9 is a flowchart illustrating an example of the operation of the estimation device 101 when predicting the health state of a target person.

[0123] The acquiring unit 140 acquires target data (S201). For example, the acquiring unit 140 acquires the target data from a terminal device.

[0124] The prediction unit 170 identifies a data group in the first distribution that corresponds to the target data (S202). For example, the prediction unit 170 identifies which data group the target data falls into in the probability density distribution related to health data in the first age group.

[0125] Then, prediction unit 170 predicts the health data of the target person when he / she reaches the age of the second age group based on the state transition probability (S203). Specifically, prediction unit 170 predicts the health data of the target person when he / she reaches the age of the second age group based on the state transition probability of the identified data group in the probability density distribution of the health data in the second age group.

[0126] Note that this operation example is merely an example, and the operation of the estimation device 101 of the present disclosure is not limited to this example.

[0127] As described above, the estimation device 101 of the third embodiment acquires health data of a target person. In this case, the first age group corresponds to the target person's age. The estimation device 101 then predicts the data group in the second distribution, which is the transition destination of the data group in the first distribution corresponding to the target person's health data based on the state transition probability, as the health data of the target person when they reach the age of the second age group. This allows the estimation device 101 to predict the target person's future health state.

[0128] <Fourth embodiment> Next, an estimation device according to a fourth embodiment will be described. In the fourth embodiment, a further example of predicting the health state of a target person based on calculated state transition probabilities will be described. In the fourth embodiment, an example in which the estimation device estimates data transitions based on health data will also be mainly described, but the target data is not limited to this example. Note that some of the content overlapping with the first, second, and third embodiments will not be described.

[0129] [Details of the estimation device 102] The estimation device 102 is a device obtained by adding additional functional units to the estimation device 101. FIG. 10 is a block diagram showing an example of the functional configuration of the estimation device 102. The estimation device 102 includes a reception unit 110, an estimation unit 120, an output unit 130, an acquisition unit 140, a classification unit 150, and a prediction unit 170. The estimation device 101 may also include a generation unit 180. The estimation device 102 may also include a storage device 190.

[0130] Like the estimation device 101, the estimation device 102 may be a device provided in a terminal device, or may be a device implemented in a server device communicatively connected to the terminal device via a wired or wireless network.

[0131] The estimation device 102 calculates state transition probabilities between stages in advance. Then, the estimation device 102 generates a learning model using the calculated state transition probabilities. The estimation device 102 also uses the generated learning model to predict future health data for the target person. In this embodiment, the stage of generating the learning model is referred to as the generation phase, and the stage of making predictions is referred to as the prediction phase.

[0132] (Generation phase) It is assumed that a data set for each stage is stored in advance in storage device 190. For example, it is assumed that a probability density distribution regarding health data for each age group is stored in storage device 190.

[0133] The receiving unit 110 receives input of a probability density distribution related to health data for each age group. In this embodiment, the probability density distribution is a distribution related to health data for each age group of 10 years. Furthermore, it is assumed that there are eight types of probability density distributions, from the teens to the 80s.

[0134] The estimation unit 120 estimates state transition probabilities between stages. For example, the estimation unit 120 estimates a state transition probability based on a transition from a probability density distribution for health data for people in their teens to a probability density distribution for health data for people in their twenties. The estimation unit 120 also estimates a state transition probability based on a transition from a probability density distribution for health data for people in their twenties to a probability density distribution for health data for people in their thirties. Similarly, the estimation unit 120 estimates a state transition probability based on a transition in data distribution for each stage based on the probability density distribution for health data between each stage. In this case, the estimation unit 120 may also estimate a state transition probability between stages that are not adjacent. For example, the estimation unit 120 may estimate a state transition probability based on a transition from a probability density distribution for health data for people in their twenties to a probability density distribution for health data for people in their fifties.

[0135] The output section 130 outputs the state transition probability regarding the transition of data between each stage. At this time, the output section 130 may store the state transition probability and the set of stages corresponding to the state transition probability in the storage device 190 in association with each other.

[0136] The generation unit 180 generates a machine learning model. Specifically, the generation unit 180 generates a prediction model that outputs data in another stage, which is a transition destination of data in one stage, based on the output state transition probability. The prediction model corresponds to a machine learning model that learns the relationship between the distribution of data in one stage and the distribution of data in another stage. Alternatively, the generation unit 180 may generate a prediction model that outputs a data group of data in another stage, which is a transition destination of a data group of data distribution in one stage, based on the output state transition probability. The generated prediction model is a machine learning model that receives age and health data as input, and outputs health data to which the input health data will transition after a predetermined period, or a group of such data.

[0137] In this way, generation unit 180 generates a machine learning model that learns the relationship between data in a stage before a transition and data in a stage after a transition based on state transition probabilities. More specifically, generation unit 180 generates a machine learning model that learns the relationship between health data in an age group before a transition and health data in an age group after a transition based on state transition probabilities. Generation unit 180 is an example of a generation means.

[0138] (Prediction phase) The acquisition unit 140 acquires target data, which is health data of a target person.

[0139] The prediction unit 170 uses a machine learning model to estimate the future health state of the target person. Specifically, the prediction unit 170 predicts the transition of target data, which is health data related to the target person. For example, the prediction unit 170 inputs the target data and the target person's age into a machine learning model. The machine learning model outputs health data for age groups above the target person's age. In other words, the health data of the target person when they reach the age after a predetermined period of time has passed is output. The prediction unit 170 predicts the health data as the target person's future health data.

[0140] For example, suppose a target person is 51 years old. Suppose health data for the target person when they reach their 70s is to be predicted. In this case, prediction unit 170 inputs, for example, information indicating that the target person is 51 years old and the target data into the machine learning model. At this time, the machine learning model outputs health data for the 70s age group, which is the transition destination of the input health data. Prediction unit 170 outputs the output health data as health data for the target person when they reach their 70s.

[0141] The machine learning model may be a model that outputs health data for target data after a specific period of time has passed. For example, the machine learning model may output health data for the target person when they reach an age group 20 years from now. The machine learning model may also be a model that outputs trends in health data up to a specific period of time has passed. For example, the machine learning model may output trends in health data for the target person until they reach an age group 20 years from now.

[0142] [Example of operation of the estimation device 102] Next, an example of the operation of the estimation device 102 will be described with reference to FIGS.

[0143] Fig. 11 is a fourth flowchart illustrating an example of the operation of estimation device 102. Specifically, Fig. 11 is a flowchart illustrating an example of the operation of estimation device 102 in the generation phase. In the operation example of Fig. 11, it is assumed that a probability density distribution related to health data for each age group is stored in advance in storage device 190.

[0144] The receiving unit 110 receives an input of a probability density distribution related to health data for each age group (S301). For example, the receiving unit 110 receives an input of a probability density distribution related to health data for each age group.

[0145] The estimation unit 120 estimates the state transition probability for each stage (S302). Specifically, the estimation unit 120 estimates the state transition probability based on the transition of the data distribution for each stage, based on the probability density distribution of the health data between the stages.

[0146] The output unit 130 outputs the state transition probability for each stage (S303). Then, the generation unit 180 generates a machine learning model based on the output state transition probabilities (S304). Specifically, the generation unit 180 generates a machine learning model that learns the relationship between data in a stage before the transition and data in a stage after the transition based on the state transition probabilities.

[0147] Fig. 12 is a fifth flowchart illustrating an example of the operation of the estimating device 102. Specifically, Fig. 12 is a flowchart illustrating an example of the operation of the estimating device 102 in the prediction phase.

[0148] The acquiring unit 140 acquires target data (S401). For example, the acquiring unit 140 acquires the target data from a terminal device.

[0149] Prediction unit 170 uses a machine learning model to predict the future health state of the target person (S402). Specifically, prediction unit 170 inputs information indicating the target person's age and the target data into the machine learning model. Health data is then output by the machine learning model. Prediction unit 170 outputs the output health data as health data for the target person when they reach the age after a predetermined period of time has passed.

[0150] Note that this operation example is merely an example, and the operation of the estimation device 102 of the present disclosure is not limited to this example.

[0151] [Variation 2] In the present disclosure, examples have been described in which the estimation device estimates data transitions based on health data. That is, examples have been described in which the estimation device is used in the fields of healthcare or medicine. However, examples in which the estimation device is applied are not limited to these. For example, the estimation device may also be applied to estimating state transitions of various machines.

[0152] For example, when measurement data is acquired for the operating state of a machine, the estimation device may accept the measurement data for each state based on aging, from a state in which the machine is operating normally to a state in which the machine is broken, as a data set for each stage. The estimation device may also estimate state transition probabilities based on the distribution of the measurement data between each state. The estimation device may then output state transition probabilities for transitions between each state.

[0153] <Example of hardware configuration of estimation device> The hardware constituting the estimation devices of the first, second, third, and fourth embodiments described above will now be described. FIG. 13 is a block diagram showing an example of the hardware configuration of a computer device constituting the estimation device in each embodiment. The estimation device and estimation method described in each embodiment and each modified example are realized in a computer device 90. For example, the estimation device and the like described in each embodiment and each modified example may have the hardware configuration shown in FIG. 13.

[0154] 13, a computer device 90 includes a processor 91, a RAM (Random Access Memory) 92, a ROM (Read Only Memory) 93, a storage device 94, an input / output interface 95, a bus 96, and a drive device 97. Note that the estimation device and the like may be realized by a plurality of electric circuits.

[0155] The storage device 94 stores a program (computer program) 98. The processor 91 executes the program 98 of the present estimation device using the RAM 92. Specifically, for example, the program 98 includes a program that causes a computer to execute processes shown in FIGS. 2, 6, 9, 11, and 12, etc. The processor 91 executes the program 98 to realize the functions of each component of the present estimation device. The program 98 may be stored in the ROM 93. Alternatively, the program 98 may be recorded on the recording medium 80 and read out using the drive device 97, or may be transmitted to the computer device 90 from an external device (not shown) via a network (not shown).

[0156] The input / output interface 95 exchanges data with peripheral devices (such as a keyboard, a mouse, and a display device) 99. The input / output interface 95 functions as a means for acquiring or outputting data. The bus 96 connects each component.

[0157] There are various variations in the method for realizing the estimation device. For example, each component included in the estimation device can be realized as a dedicated device. Furthermore, the estimation device can be realized based on a combination of multiple devices.

[0158] The scope of each embodiment also includes a processing method for recording a program for realizing each configuration of the function of each embodiment on a recording medium, reading the program recorded on the recording medium as code, and executing it on a computer. In other words, a computer-readable recording medium is also included in the scope of each embodiment. Furthermore, a recording medium on which the above-mentioned program is recorded and the program itself are also included in each embodiment.

[0159] The recording medium may be, but is not limited to, a floppy disk, a hard disk, an optical disk, a magneto-optical disk, a CD (Compact Disc)-ROM, a magnetic tape, a non-volatile memory card, or a ROM. The programs recorded on the recording medium are not limited to standalone programs that execute processes, but also include programs that run on an OS (Operating System) in cooperation with other software and functions of an expansion board.

[0160] Although the present invention has been described above with reference to the embodiments, the present invention is not limited to the above-described embodiments. Various modifications that can be understood by those skilled in the art can be made to the configuration and details of the present invention within the scope of the present invention.

[0161] Furthermore, the above-described embodiments and modifications can be combined as appropriate.

[0162] Some or all of the above-described embodiments can be described as, but are not limited to, the following supplementary notes.

[0163] <Additional Notes> [Appendix 1] A receiving means for receiving input of a data set for each stage; an estimation means for estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; and an output means for outputting the estimated state transition probability. Estimation device.

[0164] [Appendix 2] the receiving means receives health data, which is data on the health of each of a plurality of people at a predetermined time point, for each age group, as a data set for each stage; the first distribution is a distribution of health data in a first age group; The second distribution is a distribution of health data in a second age group that is an age group after the first age group, 10. The estimation apparatus of claim 1.

[0165] [Appendix 3] the estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on a cost function such that a cost of the health data of the first distribution transitioning in a worsening direction is smaller than a cost of the health data of the first distribution transitioning in an improving direction, 10. The estimation device of claim 2.

[0166] [Appendix 4] The health data includes values ​​of a plurality of health-related items for each of a plurality of persons; the first distribution and the second distribution are multidimensional distributions each having a plurality of health-related items as an axis, The estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on a cost function in which the cost of the health data of the first distribution transitioning varies depending on the magnitude of the value corresponding to a specific axis in the transition of the health data from the first distribution to the second distribution. 10. The estimation device of claim 2.

[0167] [Appendix 5] a design means for designing a cost function that calculates a cost determined for each data item based on a change in data distribution between a data item in a stage before the transition and a data item in a stage after the transition; the estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on the generated cost function; 10. The estimation apparatus of claim 1.

[0168] [Appendix 6] the design means designs a cost function including a function whose parameters are adjusted in accordance with changes in the corresponding data in the stage before the transition and the corresponding data in the stage after the transition, based on correspondence data in which the correspondence of the transition is known; 6. The estimation device of claim 5.

[0169] [Appendix 7] the estimation means estimates the state transition probability using an optimal transportation algorithm that calculates a pair of data before transportation and data at a transportation destination, optimizing the cost of transportation from the first distribution to the second distribution; 10. The estimation apparatus of claim 1.

[0170] [Appendix 8] A generating means for generating a machine learning model; and a prediction means for predicting a transition of the target data, which is health data relating to the target person; the estimation means estimates a state transition probability based on a transition in the distribution of health data for each age group; the output means outputs the state transition probability for each age group; the generation means generates the machine learning model that learns the relationship between the health data in the age group before the transition and the health data in the age group after the transition based on state transition probabilities; The prediction means predicts a transition of target data by using the machine learning model. 10. The estimation device of claim 2.

[0171] [Appendix 9] Accepts input of a dataset for each stage, estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; Output the estimated state transition probabilities. Estimation method.

[0172] [Appendix 10] A process for accepting input of a dataset for each stage; a process of estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; and outputting the estimated state transition probabilities. program.

[0173] Furthermore, some or all of the configurations described in Supplementary Notes 2 to 8, which are dependent on Supplementary Note 1, may also be dependent on Supplementary Notes 9 and 10 in the same dependent relationship as Supplementary Notes 2 to 8. Furthermore, within the scope of each of the above-mentioned embodiments, some or all of the configurations described as Supplements may be made dependent on various hardware, software, various recording means for recording software, or systems. [Explanation of symbols]

[0174] 100, 101, 102 Estimator 110 Reception 120 Estimation part 130 Output section 140 Acquisition Department 150 Classification Department 160 Design Department 170 Prediction Department 180 Generation part 190 Storage device

Claims

1. A receiving means for receiving input of a data set for each stage; an estimation means for estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; and an output means for outputting the estimated state transition probability. Estimation device.

2. the receiving means receives health data, which is data on the health of each of a plurality of people at a predetermined time point, for each age group, as a data set for each stage; the first distribution is a distribution of health data in a first age group; The second distribution is a distribution of health data in a second age group that is an age group subsequent to the first age group. The estimation device according to claim 1 .

3. the estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on a cost function such that a cost of the health data of the first distribution transitioning in a worsening direction is smaller than a cost of the health data of the first distribution transitioning in an improving direction, The estimation device according to claim 2 .

4. The health data includes values ​​of a plurality of health-related items for each of a plurality of persons; the first distribution and the second distribution are multidimensional distributions each having a plurality of health-related items as an axis, the estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on a cost function in which the cost of the health data of the first distribution transitioning varies depending on the magnitude of the value corresponding to a specific axis in the transition of the health data from the first distribution to the second distribution; The estimation device according to claim 2 .

5. a design means for designing a cost function that calculates a cost determined for each data item based on a change in data distribution between a data item in a stage before the transition and a data item in a stage after the transition; the estimation means estimates a state transition probability based on a transition from the first distribution to the second distribution based on the generated cost function; The estimation device according to claim 1 .

6. the design means designs a cost function including a function whose parameters are adjusted in accordance with changes in the corresponding data in the stage before the transition and the corresponding data in the stage after the transition, based on correspondence data in which the correspondence of the transition is known; The estimation device according to claim 5 .

7. the estimation means estimates the state transition probability using an optimal transportation algorithm that calculates a pair of data before transportation and data at a transportation destination, optimizing the cost of transportation from the first distribution to the second distribution; The estimation device according to claim 1 .

8. A generating means for generating a machine learning model; and a prediction means for predicting a transition of the target data, which is health data relating to the target person; the estimation means estimates a state transition probability based on a transition in the distribution of health data for each age group; the output means outputs the state transition probability for each age group; the generation means generates the machine learning model that learns the relationship between the health data in the age group before the transition and the health data in the age group after the transition based on state transition probabilities; The prediction means predicts a transition of target data by using the machine learning model. The estimation device according to claim 2 .

9. Accepts input of a dataset for each stage, estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; Output the estimated state transition probabilities. Estimation method.

10. A process for accepting input of a dataset for each stage; a process of estimating a state transition probability based on a transition from a first distribution, which is a distribution of data in a first stage, to a second distribution, which is a distribution of data in a second stage, based on a cost required for transition to a transition destination, which is determined for each data in the first distribution; and outputting the estimated state transition probabilities. program.

Images