Feature validity evaluation method, device, equipment and storage medium

Abstract

The present invention discloses a feature validity evaluation method, device, equipment and storage medium. The feature validity evaluation method is applied to the risk control model of the Medical Insurance Bureau. Specifically, a first valid feature data set is obtained through a tree structure model, a second valid feature data set is obtained through a deep learning neural network model, and a third valid feature data set is obtained through a target business scenario. Then, the union of the first valid feature data set, the second valid feature data set and the third valid feature data set is used as the target valid feature data set, so that high-quality features can be selected for the risk control model through multiple dimensions, thereby facilitating the intelligent risk control of big data in medical insurance.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a feature validity evaluation method, device, equipment and storage medium. Background Art

[0002] Feature engineering is an important part of the big data risk control field. The richness of features generated by feature engineering determines the quality of machine learning and deep learning models to a certain extent. Feature engineering will generate tens of thousands of different types of features. Extracting useful features from these features is an important part of feature engineering. Useless features will interfere with the training model, making the training time longer, and sometimes even leading to problems such as underfitting of the model.

[0003] At present, the feature engineering effectiveness evaluation method is largely based on manual screening by business personnel. It requires high business capabilities of business personnel, has high labor costs, and is not comprehensive when selecting effective features, which can easily lead to omissions. Some automated feature effectiveness evaluation methods are scattered and unsystematic.

[0004] Therefore, the existing technology needs to be improved. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a feature validity evaluation method, device, equipment and storage medium in response to the above-mentioned defects of the prior art, aiming to solve the technical problem that the feature validity evaluation method in the prior art is time-consuming, labor-intensive and prone to omissions when selecting effective features.

[0006] In a first aspect, the present application provides a method for evaluating feature effectiveness, the method comprising:

[0007] The importance of each feature data in the initial feature data set is obtained by using a tree structure model, and the feature data whose importance is greater than an importance threshold is taken as valid feature data to obtain a first valid feature data set;

[0008] Inputting the initial feature data set into a deep learning neural network model to obtain a first result, acquiring a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and inputting the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when an error between the first result and the second result is greater than an error threshold, using the second subset of initial feature data as valid feature data to obtain a second valid feature data set, wherein the initial feature data set includes the first subset of initial feature data and the second subset of initial feature data;

[0009] The initial feature data set is obtained in the business dimension, and feature data in the target business dimension related to the target business scenario in the business dimension is obtained as valid feature data to obtain a third valid feature data set, wherein the business dimension includes one or more of a personnel number dimension, a hospital dimension, a medical number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features;

[0010] The union of the first valid feature data set, the second valid feature data set, and the third valid feature data set is taken as the target valid feature data set.

[0011] Optionally, the tree structure model includes a random forest model, a composite tree model xgboost and a decision tree model lightgbm, and the tree structure model is used to obtain the importance of each feature data in the initial feature data set, and the feature data whose importance is greater than the importance threshold is used as the valid feature data to obtain the first valid feature data set, including:

[0012] Using a random forest model to obtain a first importance of each feature data in the initial feature data set, and taking feature data whose first importance is greater than an importance threshold as valid feature data to obtain a fourth valid feature data set;

[0013] Using the composite tree model xgboost to obtain the second importance of each feature data in the initial feature data set, and taking the feature data whose second importance is greater than the importance threshold as valid feature data to obtain a fifth valid feature data set;

[0014] Using the decision tree model lightgbm, obtain the third importance of each feature data in the initial feature data set, and take the feature data whose third importance is greater than the importance threshold as valid feature data to obtain a sixth valid feature data set;

[0015] The intersection of the fourth valid feature data set, the fifth valid feature data set and the sixth valid feature data set is determined as the first valid feature data set, or the union of the fourth valid feature data set, the fifth valid feature data set and the sixth valid feature data set is determined as the first valid feature data set.

[0016] Optionally, the deep learning neural network model includes three hidden layers, the initial feature data set is input into the deep learning neural network model to obtain a first result, a first subset of initial feature data and a second subset of initial feature data are obtained based on the initial feature data set, and the first subset of initial feature data is input into the deep learning neural network model to obtain a second result, and when an error between the first result and the second result is greater than an error threshold, the second subset of initial feature data is used as valid feature data to obtain a second valid feature data set, including:

[0017] Inputting the initial feature data set into a deep learning neural network model to obtain a first result;

[0018] Selecting three quarters of the initial feature data as the first subset of the initial feature data, and selecting one quarter of the initial feature data as the second subset of the initial feature data;

[0019] Inputting the first subset of the initial feature data into the deep learning neural network model to obtain a second result;

[0020] When the error between the first result and the second result is greater than an error threshold, reselecting feature data with a smaller proportion than the last selection from the initial feature data as the first subset of the initial feature data, and reselecting feature data with a greater than or equal proportion than the last selection from the initial feature data as the second subset of the initial feature data;

[0021] Re-inputting the first subset of the initial feature data into the deep learning neural network model to obtain a new first result;

[0022] When the error between the first result and the second result is greater than an error threshold, repeatedly selecting feature data less than the last selected ratio from the initial feature data as the first subset of the initial feature data, and selecting feature data greater than or equal to the last selected ratio from the initial feature data as the second subset of the initial feature data, until when the error between the first result and the second result is less than or equal to the error threshold, the second subset of the initial feature data selected last time is used as the valid feature data to obtain a second valid feature data set.

[0023] Optionally, after re-inputting the first subset of the initial feature data into the deep learning neural network model to obtain a new first result, the method further includes:

[0024] When the error between the first result and the second result is less than or equal to an error threshold, the second subset of initial feature data is removed from the initial feature data subset to obtain a new data set, so as to update the initial feature data subset.

[0025] Optionally, taking the union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set includes:

[0026] Obtaining an intersection of the first valid feature data set, the second valid feature data set, and the third valid feature data set;

[0027] The intersection is deduplicated to obtain the target valid feature data set.

[0028] Optionally, the target business scenarios include frequent hospitalizations in low-level hospitals, frequent patients with multiple diseases in low-level hospitals, prescriptions in groups, and outpatient seniority monitoring. The feature data on the target business dimension related to the target business scenario in the business dimension is obtained as valid feature data to obtain a third valid feature data set based on the target business scenario, including:

[0029] In the case where the target business scenario includes hospitalization in low-level hospitals, a third valid feature data set is obtained based on the feature data on the maximum annual hospitalization cost in the hospital dimension, the number of non-tertiary hospital hospitalization institutions in the department dimension, the non-tertiary hospitals in the department dimension, the number of tertiary hospital hospitalizations in the medical number dimension, and the number of times the hospitalization cost in the hospital dimension is less than the average hospitalization cost of the insured persons in the city as valid feature data;

[0030] In the case where the target business scenario includes frequent patients with multiple diseases in the low-level hospital, the feature data based on the number of hospitalizations in the primary hospital in the medical number dimension, the number of patient diseases in the personnel number dimension, the ratio of patient hospitalization to outpatient care in the insured unit dimension, the total hospitalization expenses of patients in the multi-dimensional features, the average number of items per hospitalization in the multi-dimensional features, the number of primary hospitals visited in the multi-dimensional features, the use of auxiliary drugs in the multi-dimensional features, and the distance from the primary hospital to the hospital in the multi-dimensional features are used as valid feature data to obtain a third valid feature data set;

[0031] In the case where the target business scenario includes prescribing drugs in groups, a third valid feature data set is obtained based on whether there is a suspected violation of prescribing drugs in groups in the multi-dimensional features, the number of prescribing drugs in groups in the past year in the hospital dimension, the average interval between each violation in the physician dimension, and feature data on the medical institution data in the multi-dimensional features as valid feature data;

[0032] In the case where the target business scenario includes outpatient chronic disease qualification monitoring, a third valid feature data set is obtained based on the feature data of the long distance to the first-level inpatient hospital, the necessary medication costs in the medical number dimension, the number of necessary medication pick-up times in the medical number dimension, the time interval for medication pick-up in the medical number dimension, the total medication costs in the medical number dimension, the total number of medication pick-up times in the medical number dimension, and other medical types in the multi-dimensional features as valid feature data.

[0033] Optionally, after performing deduplication processing on the intersection to obtain the target valid feature data set, the method further includes:

[0034] The target effective feature set is subjected to derivative feature processing to obtain a derived effective feature data set, and the derived effective feature data set is merged into the target effective feature data set to obtain an updated target effective feature data set.

[0035] In a second aspect, the present application provides a feature validity evaluation device, comprising:

[0036] A first acquisition module 1 is used to obtain the importance of each feature data in the initial feature data set by using a tree structure model, and take the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set;

[0037] A second acquisition module 2 is used to input the initial feature data set into the deep learning neural network model to obtain a first result, obtain a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and input the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when the error between the first result and the second result is greater than an error threshold, use the second subset of initial feature data as valid feature data to obtain a second valid feature data set, wherein the initial feature data set includes the first subset of initial feature data and the second subset of initial feature data;

[0038] A third acquisition module 3 is used to acquire the initial feature data set in the business dimension, and acquire the feature data in the target business dimension related to the target business scenario in the business dimension as valid feature data to obtain a third valid feature data set based on the target business scenario, wherein the business dimension includes one or more of a personnel number dimension, a hospital dimension, a medical number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features;

[0039] The processing module 4 is configured to take the union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set.

[0040] In a third aspect, the present application provides a computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the feature validity evaluation method as described in any of the above technical solutions is implemented.

[0041] In a fourth aspect, the present application provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the feature validity evaluation method as described in any of the above technical solutions.

[0042] Beneficial effect: The present invention provides a method for evaluating feature effectiveness, the method comprising: using a tree structure model to obtain the importance of each feature data in an initial feature data set, taking the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set; inputting the initial feature data set into a deep learning neural network model to obtain a first result, obtaining a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and inputting the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when the error between the first result and the second result is greater than an error threshold, taking the second subset of initial feature data as valid feature data. A second valid feature data set is obtained, wherein the initial feature data set includes the first subset of the initial feature data and the second subset of the initial feature data; the initial feature data set is obtained in the business dimension, and the feature data in the target business dimension related to the target business scenario in the business dimension is obtained as valid feature data to obtain a third valid feature data set, wherein the business dimension includes one or more of the personnel number dimension, hospital dimension, medical number dimension, department dimension, physician dimension, participating unit dimension and multi-dimensional features; the union of the first valid feature data set, the second valid feature data set and the third valid feature data set is used as the target valid feature data set. In this solution, the feature validity evaluation method is applied to the risk control model of the Medical Insurance Bureau. Specifically, the first valid feature data set is obtained through a tree structure model, the second valid feature data set is obtained through a deep learning neural network model, and the third valid feature data set is obtained through the target business scenario. Then, the union of the first valid feature data set, the second valid feature data set and the third valid feature data set is used as the target valid feature data set, so that high-quality features can be selected for the risk control model through multiple dimensions, which helps the intelligent risk control of medical insurance big data. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the background technology, the drawings required for use in the embodiments of the present application or the background technology will be described below.

[0044] The drawings herein are incorporated into the specification and constitute a part of the specification. These drawings illustrate embodiments consistent with the present application and are used together with the specification to illustrate the technical solution of the present application.

[0045] Figure 1 A flowchart of a feature effectiveness evaluation method for this application;

[0046] Figure 2 A schematic diagram of the principle of a feature effectiveness evaluation method for this application;

[0047] Figure 3 This is a schematic diagram of the structure of a feature effectiveness evaluation device of the present application;

[0048] Figure 4 This is a schematic diagram of the hardware structure of a feature effectiveness evaluation device for this application. DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0050] At present, the feature engineering effectiveness evaluation method is largely based on manual screening by business personnel. It requires high business capabilities of business personnel, has high labor costs, and is not comprehensive when selecting effective features, which can easily lead to omissions. Some automated feature effectiveness evaluation methods are scattered and unsystematic.

[0051] Based on this, the present application hopes to provide a solution that can solve the above-mentioned technical problems, the details of which will be described in the subsequent embodiments.

[0052] Figure 1 It is a flowchart of a feature effectiveness evaluation method, such as Figure 1 As shown, the feature validity evaluation method provided in the embodiment of the present invention is applied to the risk control model of the Medical Insurance Bureau, aiming to select high-quality features for the big data risk control model, lay a solid foundation for the model, and assist in the intelligentization of big data risk control of medical insurance.

[0053] The method can be applied to a feature validity evaluation device, which can be a terminal device, a server or other processing device, etc. The terminal device can be a user equipment (UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc.

[0054] like Figure 1 As shown, the feature effectiveness evaluation method includes:

[0055] S100, using a tree structure model to obtain the importance of each feature data in the initial feature data set, and taking the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set;

[0056] In this embodiment, the initial feature data set is a set containing multiple initial data features. Specifically, the data type of the initial feature can be numerical, categorical, time series, statistical, multidimensional, etc. The tree structure model is pre-set. When any data in the initial feature data set is input into the tree structure model, the importance of the feature data can be obtained, and the feature data with importance greater than the importance threshold can be screened out as the first valid feature data set.

[0057] Specifically, the importance threshold can be set according to actual needs. The value of the importance threshold affects the accuracy of the first small feature data set. When it is necessary to improve the accuracy of the first effective feature data set, the preset value of the importance threshold can be increased.

[0058] S200, inputting the initial feature data set into a deep learning neural network model to obtain a first result, obtaining a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and inputting the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when an error between the first result and the second result is greater than an error threshold, using the second subset of initial feature data as valid feature data to obtain a second valid feature data set, wherein the initial feature data set includes the first subset of initial feature data and the second subset of initial feature data;

[0059] In this embodiment, the deep learning neural network model is trained based on sample feature data. Specifically, the sample feature data can be numerical, categorical, time series, statistical, multi-dimensional, etc. Specifically, step S200 includes:

[0060] A. When the initial feature data in the initial feature data set is first input into the deep learning neural network, a first result is obtained;

[0061] In one embodiment, when the initial feature data set is some information related to the patient, for example, the initial feature data set includes personal information, historical medication information, historical medical information, medical insurance information, work status, and educational information, the deep learning neural network model can be a neural network model for recommending medication to patients, so that the first result can be the medication recommendation result for the first patient.

[0062] B. Then the above initial feature data set is divided into a first subset of initial feature data and a second subset of initial feature data, wherein the first subset of initial feature data and the second subset of initial feature data are both part of the initial feature data set.

[0063] In one embodiment, the first subset of initial features and the second subset of initial features constitute the entire set of initial features.

[0064] In a specific embodiment, when the initial feature data set includes personal information, historical medication information, historical medical information, medical insurance information, work status, and educational information, the initial feature first subset includes educational information and work status, and the initial feature data second subset includes personal information, historical medication information, historical medical information, and medical insurance information, so that the initial feature first subset and the initial feature second subset constitute the complete initial feature data set.

[0065] C. Input the first subset of the initial feature data into the deep learning neural network model to obtain a second result; when the error between the first result and the second result is greater than an error threshold, use the second subset of the initial feature data as valid feature data to obtain a second valid feature data set.

[0066] Since the first result is the output result of the deep learning neural network model under the initial feature data set, after removing the second subset of the initial feature data from the initial feature data set, only the first subset of the initial feature data is input into the deep learning neural network model to obtain the second result, and by comparing the first result with the second result, the validity of the second subset of the initial feature data can be reflected. Because if the output result of the deep learning neural network model is greatly affected after removing the second subset of the initial feature data, it means that this part of the data has a greater impact on the output result, which means that the data is valid; when the error between the first result and the second result is less than the error threshold, it means that removing the second subset of the initial feature data has little effect on the output structure of the deep learning neural network model, so the second subset of the initial feature data is determined to be irrelevant data, and the first subset of the initial feature data is determined to be the second valid feature set.

[0067] In a specific embodiment, when the initial feature data set includes personal information, historical medication information, historical medical information, medical insurance information, work status, and educational information, the first result is that the initial feature data set is input into the deep learning neural network model, and the first patient recommended medication result is medication A. When the initial feature first subset includes educational information and work status, the initial feature data second subset includes personal information, historical medication information, historical medical information, and medical insurance information. The deep learning neural network model is a neural network model for recommending medication to patients, so when only educational information and work status are input into the deep learning neural network model, the second patient recommended medication result is also medication A. Further, by comparing the first patient recommended medication result with the second patient recommended medication result, a conclusion is obtained that the error is less than the threshold, which means that the first subset of the initial feature data, namely educational information and work status, is valid information;

[0068] In another embodiment, when the initial feature data set includes personal information, historical medication information, historical medical information, medical insurance information, work status, and educational information, the first result is that the first patient recommended medication result output by inputting the initial feature data set into the deep learning neural network model is medication A, and when the initial feature first subset includes educational information and work status, the initial feature data second subset includes personal information, historical medication information, historical medical information, and medical insurance information. The deep learning neural network model is a neural network model for recommending medication to patients, so when only educational information and work status are input into the deep learning neural network model, the second patient recommended medication result is medication B. Further, by comparing the first patient recommended medication result with the second patient recommended medication result, a conclusion is obtained that the error is greater than the threshold, which means that the second subset of the initial feature data, namely personal information, historical medication information, historical medical information, and medical insurance information, is valid information.

[0069] S300, obtaining the initial feature data set in the business dimension, obtaining feature data in the target business dimension related to the target business scenario in the business dimension as valid feature data to obtain a third valid feature data set, wherein the business dimension includes one or more of a personnel number dimension, a hospital dimension, a medical number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features;

[0070] In this embodiment, the business dimension refers to the features related to the target business scenario, and the features related to the target business scenario can be the personnel number dimension, hospital dimension, medical number dimension, department dimension, physician dimension, insurance unit dimension, etc. A variety of features can be generated based on these business dimensions, and manual personnel can select some features according to actual needs as the third effective feature set. The manual screening method is more targeted and can complete the efficacy features to avoid missing effective features.

[0071] S400: Taking the union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set.

[0072] In this embodiment, on the one hand, a first effective feature data set is obtained through a tree structure model, on the second hand, a second effective feature data set is obtained through a deep learning neural network model, and on the third hand, the third effective feature data set is completed through the target business scenario, and then the union of the first effective feature data set, the second effective feature data set, and the third effective feature data set is used as the target effective feature data set, so that high-quality features can be selected for the risk control model through multiple dimensions. The feature validity evaluation method is applied to the risk control model of the Medical Insurance Bureau to obtain a multi-dimensional effective feature set, which helps the intelligent risk control of big data in medical insurance.

[0073] As an optional implementation, the tree structure model includes a random forest model, a composite tree model xgboost and a decision tree model lightgbm, and the step S100 includes:

[0074] S102, using a random forest model to obtain a first importance of each feature data in the initial feature data set, and taking the feature data whose first importance is greater than an importance threshold as valid feature data to obtain a fourth valid feature data set;

[0075] Specifically, random forest refers to a classifier that uses multiple trees to train and predict initial feature data. After the random forest model is trained, the initial feature data is input to obtain the first importance result of the initial feature data. When the first importance is greater than the preset importance threshold, the initial feature data is used as the fourth effective feature data. When each data in the initial feature data set is input into the random forest model in turn, the fourth effective feature data set can be obtained.

[0076] S104, using the composite tree model xgboost to obtain the second importance of each feature data in the initial feature data set, and taking the feature data whose second importance is greater than the importance threshold as valid feature data to obtain a fifth valid feature data set;

[0077] In this embodiment, after the composite tree model is trained, the initial feature data is input into the composite number model to obtain the second importance result of the initial feature data. When the second importance is greater than the preset importance threshold, the initial feature data is used as the fifth effective feature data. After each data in the initial feature data set is input into the composite tree model in sequence, the fifth effective feature data set can be obtained.

[0078] S106, using the decision tree model lightgbm to obtain the third importance of each feature data in the initial feature data set, and taking the feature data whose third importance is greater than the importance threshold as valid feature data to obtain a sixth valid feature data set;

[0079] In this embodiment, after training the decision tree model lightgbm, the initial feature data is input into the decision tree model to obtain the third importance result of the initial feature data. When the third importance is greater than the preset importance threshold, the initial feature data is used as the sixth effective feature data. When each data in the initial feature data set is input into the decision tree model in turn, the fifth effective feature data set can be obtained.

[0080] S108. Determine the intersection of the fourth valid feature data set, the fifth valid feature data set, and the sixth valid feature data set as the first valid feature data set, or determine the union of the fourth valid feature data set, the fifth valid feature data set, and the sixth valid feature data set as the first valid feature data set.

[0081] In this embodiment, three tree structure models, random forest, xgboost, and lightgbm, are selected. Since the algorithms of each model are inconsistent, the importance of the same initial feature data to the three tree structure models may be different. Therefore, after obtaining the fourth effective feature data set, the fifth effective feature data set, and the sixth effective feature data set, when the intersection of the fourth effective feature data set, the fifth effective feature data set, and the sixth effective feature data set is used as the first effective feature data set, the comprehensiveness of the first effective feature data set can be increased to avoid missing effective features. When the union of the fifth effective feature data set and the sixth effective feature data set is used as the first effective feature data set, the accuracy of the first effective feature data set can be increased, which is convenient for selecting effective features with higher quality and stronger correlation.

[0082] As an optional implementation, the deep learning neural network model includes three hidden layers, and the step S200 includes:

[0083] S202, inputting the initial feature data set into a deep learning neural network model to obtain a first result;

[0084] In one embodiment, when the initial feature data set is some information related to the patient, for example, the initial feature data set includes personal information, historical medication information, historical medical information, medical insurance information, work status, and educational information, the deep learning neural network model can be a neural network model for recommending medication to the patient, so that the first result can be the first result of recommending medication to the patient.

[0085] S204, selecting three quarters of the initial feature data as the first subset of the initial feature data, and selecting one quarter of the initial feature data as the second subset of the initial feature data;

[0086] In one embodiment, when the initial feature data set includes personal information, historical medication information, historical medical information, and medical insurance information, the first subset of the initial feature data is personal information, historical medication information, and historical medical information, and the second subset of the initial feature data is medical insurance information.

[0087] S206, inputting the first subset of the initial feature data into the deep learning neural network model to obtain a second result;

[0088] In one embodiment, a first subset of the initial feature data, namely personal information, historical medication information, and historical medical information, is input into a deep learning neural network model to obtain a second result, which can be a second result of recommending medication for the patient.

[0089] S208. When the error between the first result and the second result is greater than an error threshold, reselect feature data with a smaller ratio than the last selection from the initial feature data as the first subset of the initial feature data, and reselect feature data with a ratio greater than or equal to the last selection from the initial feature data as the second subset of the initial feature data;

[0090] When the error between the first result and the second result is greater than the error threshold, it means that the second subset of the initial feature data affects the error, that is, the second subset of the initial feature data is a valid feature. At the same time, there are also some valid features in the second subset of the initial feature data. At this time, the allocation of the first subset of initial feature data and the second subset of initial feature data in the initial feature data is reselected. Specifically, the selection ratio of the reallocated new first subset of initial feature data is smaller than the ratio of the first subset of initial feature data selected for the first time, and the selection ratio of the reallocated new second subset of initial feature data is greater than or equal to the ratio of the first subset of initial feature data selected for the first time. Optionally, one half of the initial feature data is selected as the first subset of initial feature data, and one half of the initial feature data is selected as the second subset of initial feature data.

[0091] In one embodiment, when the initial feature data set includes personal information, historical medication information, historical medical information, and medical insurance information, the first subset of the initial feature data is personal information, historical medication information, and historical medical information, and the second subset of the initial feature data is medical insurance information. When the personal information, historical medication information, historical medical information, and medical insurance information are input into the deep learning neural network model, a first result C is obtained, and when the personal information, historical medication information, and historical medical information are input into the deep learning neural network model, a second result D is obtained. When the error between the first result C and the second result D is greater than the error threshold, it means that the medical insurance information affects the output result. At this time, the personal information in the original first subset of the initial feature data can be added to the second subset of the initial feature data, and the newly reallocated initial feature first subset is historical medication information and historical medical information, and the newly reallocated initial feature second subset is personal information and medical insurance information.

[0092] S210, re-inputting the first subset of initial feature data into the deep learning neural network model to obtain a new second result;

[0093] In one embodiment, when the reallocated new initial feature first subset is historical medication information and historical medical information, and the reallocated new initial feature second subset is personal information and medical insurance information, the historical medication information and historical medical information are input into the deep learning neural network model to obtain a new second result.

[0094] S212: When the error between the first result and the second result is greater than an error threshold, repeatedly selecting feature data less than a proportion selected last time from the initial feature data as the first subset of the initial feature data, and selecting feature data greater than or equal to a proportion selected last time from the initial feature data as the second subset of the initial feature data, until when the error between the first result and the second result is less than or equal to the error threshold, the second subset of the initial feature data selected last time is used as valid feature data to obtain a second valid feature data set.

[0095] Continue to compare the relationship between the error between the new first result and the second result and the error threshold. When the error between the new first result and the second result is greater than the error threshold, it means that there are still valid features in the newly selected first subset of initial feature data. Therefore, continue to execute the above steps S208 and S210 until the error between the first result and the second result is less than or equal to the error threshold. This means that there are no valid features in the first subset of initial feature data selected last time, that is, all valid features are in the second subset of initial feature data. At this time, the second subset of initial feature data selected last time is used as valid feature data to obtain a second valid feature data set.

[0096] In one embodiment, when the reallocated new initial feature first subset is historical medication information and historical medical information, and the reallocated new initial feature second subset is personal information and medical insurance information, the historical medication information and historical medical information are input into the deep learning neural network model to obtain a new second result E.

[0097] By comparing the first result C and the second result E, if the error between the first result C and the second result E is greater than the error threshold, it means that there is still valid feature data in the historical medication information and the historical medical information. At this time, the new initial feature first subset is reallocated to the historical medication information, and the new initial feature second subset is the historical medical information, personal information, and medical insurance information. The historical medication information is input into the deep learning neural network model to continue to obtain a new second result F. If the error value between the second result F and the first result C is less than the error threshold, it means that the historical medication information is invalid feature data. At this time, the historical medical information, personal information, and medical insurance information are used as the second valid feature data set.

[0098] Optionally, the step S200 further includes:

[0099] S214: When the error between the first result and the second result is less than or equal to an error threshold, remove the second subset of initial feature data from the initial feature data subset to obtain a new data set, so as to update the initial feature data subset.

[0100] In this embodiment, when the error between the first result and the second result is less than or equal to the error threshold, it means that the second subset of the initial feature data does not affect the judgment result of the deep learning neural network model, that is, the second subset of the initial feature data is an unimportant feature. At this time, the second subset of the initial feature data is removed from the initial feature data subset to obtain a new initial feature data subset.

[0101] As an optional implementation, the step S400 includes:

[0102] S402, obtaining the intersection of the first valid feature data set, the second valid feature data set, and the third valid feature data set;

[0103] In this embodiment, in order to further screen out more valuable valid feature data sets, after obtaining the first valid feature data set, the second valid feature data set, and the third valid feature data set, the first valid feature data set, the second valid feature data set, and the third valid feature data set are subjected to intersection processing.

[0104] S404: De-duplicate the intersection to obtain the target valid feature data set.

[0105] After obtaining the intersection of the first valid feature data set, the second valid feature data set, and the third valid feature data, since the same feature data may exist in the first valid feature data set, the second valid feature data set, and the third valid feature data, the intersection is deduplicated to avoid data duplication. The target valid data obtained by the deduplication process is a data set determined as valid feature data in all three ways, which can better reflect the accuracy of the valid feature data.

[0106] As an optional implementation, the step S300 includes:

[0107] S302, when the target business scenario includes hospitalization in low-level hospitals, a third valid feature data set is obtained based on the feature data of the maximum annual hospitalization cost in the hospital dimension, the number of non-tertiary hospital hospitalization institutions in the department dimension, non-tertiary hospitals in the department dimension, the number of hospitalizations in tertiary hospitals in the medical number dimension, and the number of times the hospitalization cost in the hospital dimension is less than the average hospitalization cost of the insured persons in the city as valid feature data;

[0108] In this implementation, valid feature data is combined with actual application scenarios, and valid features are selected through manual screening. Specifically, in the case of repeated hospitalizations in low-level hospitals: the maximum annual hospitalization cost, the number of non-tertiary hospital inpatient institutions, the number of hospitalizations in non-tertiary hospitals and tertiary hospitals, and the number of times the hospitalization cost is less than the average hospitalization cost of the city's insured persons are added as valid data feature sets to the third valid feature data set.

[0109] S304, when the target business scenario includes frequent patients with multiple diseases in low-level hospitals, the feature data based on the number of hospitalizations in the primary hospital in the medical number dimension, the number of patient diseases in the personnel number dimension, the ratio of inpatient to outpatient care in the insured unit dimension, the total hospitalization expenses of patients in the multi-dimensional features, the average number of items per hospitalization in the multi-dimensional features, the number of visits to primary hospitals in the multi-dimensional features, the use of auxiliary drugs in the multi-dimensional features, and the distance from the primary hospital to the hospital in the multi-dimensional features are used as valid feature data to obtain a third valid feature data set;

[0110] Specifically, in the case of frequent patients with multiple diseases in low-level hospitals, the following characteristic data can be added to the third effective feature data set through manual screening: the number of hospitalizations in primary hospitals, the large number of patient diseases in the personnel number dimension, the high proportion of patient inpatient and outpatient clinics in the insured unit dimension, the high total hospitalization costs of patients with multi-dimensional characteristics, the small number of items per hospitalization for patients in the multi-dimensional characteristics, the large number of visits to primary hospitals in the multi-dimensional characteristics, the high use of auxiliary medications in the multi-dimensional characteristics, and the long distance from the primary hospital to which the patient is hospitalized in the multi-dimensional characteristics.

[0111] S306. When the target business scenario includes prescribing drugs in groups, a third valid feature data set is obtained based on whether there is a suspected violation of prescribing drugs in groups in the multi-dimensional features, the number of prescribing drugs in groups in the past year in the hospital dimension, the average interval between each violation in the physician dimension, and the feature data on the medical institution data in the multi-dimensional features as valid feature data;

[0112] In the case of group prescribing, through manual screening, whether there is suspected violation of group prescribing in the multi-dimensional features, the number of group prescribing in the past year in the hospital dimension, the average interval between each violation in the physician dimension, and the feature data on the medical institution data in the multi-dimensional features can be added to the third valid feature data set as a valid feature data set.

[0113] S308. When the target business scenario includes outpatient chronic disease qualification monitoring, a third valid feature data set is obtained based on the feature data of the long distance to the first-level inpatient hospital, the necessary medication cost in the medical number dimension, the number of necessary medication pickup times in the medical number dimension, the time interval for medication pickup in the medical number dimension, the total medication cost in the medical number dimension, the total number of medication pickup times in the medical number dimension, and the drug cost under other medical types in the multi-dimensional features as valid feature data.

[0114] When the target business scenario includes outpatient chronic disease qualification monitoring, the feature data on the distance to the first-level inpatient hospital, the necessary medication costs in the medical number dimension, the number of necessary medication pick-ups in the medical number dimension, the time interval for picking up medications in the medical number dimension, the total medication costs in the medical number dimension, the total number of medication pick-ups in the medical number dimension, and the drug costs under other medical types in the multi-dimensional features are added as valid feature data to the third valid feature data set.

[0115] In the implementation mode, effective features are supplemented by experienced business personnel using business knowledge in various target business scenarios, thereby avoiding omission of target effective feature data that is finally obtained.

[0116] As an optional implementation manner, after performing deduplication processing on the intersection to obtain the target valid feature data set, the method further includes:

[0117] S500: Performing derivative feature processing on the target effective feature set to obtain a derived effective feature data set, and merging the derived effective feature data set into the target effective feature data set to obtain an updated target effective feature data set.

[0118] In this embodiment, after obtaining the target effective feature set, the effective features in the target effective feature set can also be derived to obtain a derived feature data set, and the derived feature data set is also incorporated into the target effective feature data set. In addition, after obtaining derived feature data through effective feature derivation, the effective features corresponding to the derived feature data can be deleted to avoid data duplication. For example, when BMI=weight÷height2, when weight and height are both effective features, BMI is also an effective feature, because it is likely that BMI, rather than height and weight, affects the result. At this time, height and weight can be deleted.

[0119] Based on the same inventive concept, Figure 3 As shown, the embodiment of the present application provides a feature validity evaluation device, comprising:

[0120] A first acquisition module 1 is used to obtain the importance of each feature data in the initial feature data set by using a tree structure model, and take the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set;

[0121] A second acquisition module 2 is used to input the initial feature data set into the deep learning neural network model to obtain a first result, obtain a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and input the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when the error between the first result and the second result is greater than an error threshold, use the second subset of initial feature data as valid feature data to obtain a second valid feature data set, wherein the initial feature data set includes the first subset of initial feature data and the second subset of initial feature data;

[0122] A third acquisition module 3 is used to acquire the initial feature data set in the business dimension, and acquire the feature data in the target business dimension related to the target business scenario in the business dimension as valid feature data to obtain a third valid feature data set based on the target business scenario, wherein the business dimension includes one or more of a personnel number dimension, a hospital dimension, a medical number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features;

[0123] The processing module 4 is configured to take the union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set.

[0124] As an optional implementation, the tree structure model includes a random forest model, a composite tree model xgboost and a decision tree model lightgbm, and the first acquisition module 1 includes:

[0125] A first sub-acquisition module is used to obtain the first importance of each feature data in the initial feature data set by using a random forest model, and take the feature data whose first importance is greater than an importance threshold as valid feature data to obtain a fourth valid feature data set;

[0126] A second sub-acquisition module is used to obtain the second importance of each feature data in the initial feature data set by using the composite tree model xgboost, and take the feature data whose second importance is greater than the importance threshold as the valid feature data to obtain a fifth valid feature data set;

[0127] A third sub-acquisition module is used to obtain the third importance of each feature data in the initial feature data set by using the decision tree model lightgbm, and take the feature data whose third importance is greater than the importance threshold as the valid feature data to obtain a sixth valid feature data set;

[0128] The first obtaining module is used to determine the intersection of the fourth valid feature data set, the fifth valid feature data set, and the sixth valid feature data set as the first valid feature data set, or to determine the union of the fourth valid feature data set, the fifth valid feature data set, and the sixth valid feature data set as the first valid feature data set.

[0129] As an optional implementation, the deep learning neural network model includes three hidden layers, and the second acquisition module 2 includes:

[0130] A first input module, used to input the initial feature data set into a deep learning neural network model to obtain a first result;

[0131] A first selection module, configured to select three quarters of the initial feature data as a first subset of the initial feature data, and select one quarter of the initial feature data as a second subset of the initial feature data;

[0132] A second input module, inputting the first subset of the initial feature data into the deep learning neural network model to obtain a second result;

[0133] a second selection module, for reselecting feature data less than the last selected ratio from the initial feature data as the first subset of the initial feature data, and for reselecting feature data greater than or equal to the last selected ratio from the initial feature data as the second subset of the initial feature data when the error between the first result and the second result is greater than an error threshold; a third input module, for re-inputting the first subset of the initial feature data into the deep learning neural network model to obtain a new first result;

[0134] The second obtaining module is used to repeatedly select feature data less than the last selected ratio from the initial feature data as the first subset of the initial feature data, and select feature data greater than or equal to the last selected ratio from the initial feature data as the second subset of the initial feature data when the error between the first result and the second result is greater than the error threshold, until the second subset of the initial feature data selected last time is used as the valid feature data to obtain a second valid feature data set when the error between the first result and the second result is less than or equal to the error threshold.

[0135] As an optional implementation, the second acquisition module 2 further includes:

[0136] An updating module is used to remove the second subset of initial feature data from the initial feature data subset to obtain a new data set to update the initial feature data subset when the error between the first result and the second result is less than or equal to an error threshold.

[0137] As an optional implementation, the processing module 4 includes:

[0138] A fourth acquisition module, used to acquire an intersection of the first valid feature data set, the second valid feature data set, and the third valid feature data set;

[0139] The third module performs deduplication processing on the intersection to obtain the target valid feature data set.

[0140] As an optional implementation, the third acquisition module 3 includes:

[0141] The fourth obtaining module is used for obtaining a third valid feature data set based on the feature data on the maximum annual hospitalization cost in the hospital dimension, the number of non-tertiary hospital hospitalization institutions in the department dimension, the non-tertiary hospitals in the department dimension, the number of tertiary hospital hospitalizations in the medical number dimension, and the number of times the hospitalization cost in the hospital dimension is less than the average hospitalization cost of the insured persons in the city as valid feature data when the target business scenario includes hospitalization in the low-level hospital.

[0142] A fifth obtaining module is used for obtaining a third valid feature data set based on the feature data of the number of hospitalizations in the primary hospital in the medical number dimension, the number of patient diseases in the personnel number dimension, the ratio of inpatient to outpatient care in the insured unit dimension, the total hospitalization expenses of patients in the multi-dimensional features, the average number of items per hospitalization in the multi-dimensional features, the number of visits to the primary hospital in the multi-dimensional features, the use of auxiliary drugs in the multi-dimensional features, and the distance from the primary hospital to the hospital in the multi-dimensional features as valid feature data when the target business scenario includes frequent customers with multiple diseases in the low-level hospital;

[0143] In the case where the target business scenario includes prescribing drugs in groups, a third valid feature data set is obtained based on whether there is a suspected violation of prescribing drugs in groups in the multi-dimensional features, the number of prescribing drugs in groups in the past year in the hospital dimension, the average interval between each violation in the physician dimension, and feature data on the medical institution data in the multi-dimensional features as valid feature data;

[0144] The sixth obtaining module is used to obtain a third valid feature data set based on the feature data of the long distance to the first-level inpatient hospital, the necessary medication cost in the medical number dimension, the number of necessary medication pick-up times in the medical number dimension, the time interval for medication pick-up in the medical number dimension, the total medication cost in the medical number dimension, the total number of medication pick-up times in the medical number dimension, and the drug cost under other medical types in the multi-dimensional features as valid feature data, when the target business scenario includes outpatient chronic disease qualification monitoring.

[0145] As an optional implementation, the device further includes:

[0146] The derivation module is used to perform derivative feature processing on the target effective feature set to obtain a derived effective feature data set, and merge the derived effective feature data set into the target effective feature data set to obtain an updated target effective feature data set.

[0147] Specifically, when the feature validity evaluation device is executed, it implements the steps of the feature validity evaluation method of the present invention. The specific steps of the feature validity evaluation method and the corresponding beneficial effects have been described in detail above and will not be repeated here.

[0148] Based on the same inventive concept, an embodiment of the present invention provides a computer device, such as Figure 4 As shown, it includes a memory 22, a processor 21, an input device 23, and an output device 24. The processor 21, the memory 22, the input device 23, and the output device 24 are coupled through a connector, and the connector includes various interfaces, transmission lines, or buses, etc., which are not limited in the embodiments of the present application. It should be understood that in each embodiment of the present application, coupling refers to mutual connection in a specific manner, including direct connection or indirect connection through other devices, for example, through various interfaces, transmission lines, buses, etc.

[0149] The computer device may be a laptop, tablet computer, desktop computer, mobile phone or workstation.

[0150] The processor 21 may be one or more graphics processing units (GPUs). When the processor 21 is a GPU, the GPU may be a single-core GPU or a multi-core GPU. Optionally, the processor 21 may be a processor group consisting of multiple GPUs, and the multiple processors are coupled to each other via one or more buses. Optionally, the processor may also be other types of processors, etc., which are not limited in the embodiments of the present application.

[0151] The memory 22 can be used to store computer program instructions and various computer program codes including the program code for executing the program code of the present application. Optionally, the memory includes but is not limited to random access memory (RAM), read-only memory (ROM), erasable programmable read only memory (EPROM), or portable read only memory (CD-ROM), which is used for related instructions and data.

[0152] The input device 23 is used to input data and / or signals, and the output device 24 is used to output data and / or signals. The input device 23 and the output device 24 can be independent devices or an integrated device.

[0153] It can be understood that in the embodiment of the present application, the memory 22 can be used not only to store relevant instructions, but also to store relevant data. For example, the memory 22 can be used to store target feature data obtained through the input device 23, or the memory 22 can also be used to store comparison results obtained by the processor 21, etc. The embodiment of the present application does not limit the specific data stored in the memory.

[0154] Understandably, Figure 3 Only a simplified design of a vehicle identification device is shown. In practical applications, the vehicle identification device may also include other necessary components, including but not limited to any number of input / output devices, processors, memories, etc., and all vehicle identification devices that can implement the embodiments of the present application are within the protection scope of the present application.

[0155] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0156] Based on the same inventive concept, an embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the feature validity evaluation method as described in any of the above technical solutions is implemented.

[0157] Computer-readable storage media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules or other data. Computer-readable storage media may include: read-only memory (ROM), random access memory (RAM), solid-state drive (SSD) or optical disk, etc. Among them, random access memory may include resistance random access memory (ReRAM) and dynamic random access memory (DRAM).

[0158] In summary, the present invention provides a method for evaluating feature effectiveness, such as Figure 2As shown, the method includes: using a tree structure model to obtain the importance of each feature data in an initial feature data set, taking the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set; inputting the initial feature data set into a deep learning neural network model to obtain a first result, obtaining a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and inputting the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when the error between the first result and the second result is greater than an error threshold, taking the second subset of initial feature data as valid feature data to obtain a second valid feature data. The initial feature data set includes the first subset of the initial feature data and the second subset of the initial feature data; the initial feature data set is obtained in the business dimension, and the feature data in the target business dimension related to the target business scenario in the business dimension is obtained as effective feature data to obtain a third effective feature data set, wherein the business dimension includes one or more of the personnel number dimension, hospital dimension, medical number dimension, department dimension, physician dimension, participating unit dimension and multi-dimensional features; the union of the first effective feature data set, the second effective feature data set and the third effective feature data set is used as the target effective feature data set. In this solution, the feature validity evaluation method is applied to the risk control model of the Medical Insurance Bureau. Specifically, the first effective feature data set is obtained through a tree structure model, the second effective feature data set is obtained through a deep learning neural network model, and the third effective feature data set is obtained through the target business scenario. Then, the union of the first effective feature data set, the second effective feature data set and the third effective feature data set is used as the target effective feature data set, so that high-quality features can be selected for the risk control model through multiple dimensions, which helps the intelligent risk control of medical insurance big data.

[0159] The technical principle of the present invention is described above in conjunction with specific embodiments. These descriptions are only for explaining the principle of the present invention and cannot be interpreted as limiting the scope of protection of the present invention in any way. Based on the explanations herein, those skilled in the art can associate other specific implementations of the present invention without paying creative labor, and these methods will fall within the scope of protection of the present invention.

Claims

1. A feature effectiveness evaluation method, characterized in that: The method comprises: The importance of each feature data in the initial feature data set is obtained by using a tree structure model, and the feature data whose importance is greater than an importance threshold is taken as valid feature data to obtain a first valid feature data set; Inputting the initial feature data set into a deep learning neural network model to obtain a first result, wherein the deep learning neural network model includes three hidden layers; selecting three quarters of the initial feature data set as a first subset of initial feature data, and selecting one quarter of the initial feature data set as a second subset of initial feature data; inputting the first subset of initial feature data into the deep learning neural network model to obtain a second result; When the error between the first result and the second result is greater than an error threshold, reselecting feature data with a smaller proportion than the last selection from the initial feature data set as the first subset of the initial feature data, and reselecting feature data with a proportion greater than or equal to the last selection from the initial feature data set as the second subset of the initial feature data; Re-inputting the first subset of the initial feature data into the deep learning neural network model to obtain a new first result; when the error between the first result and the second result is greater than the error threshold, repeatedly selecting feature data less than the last selected ratio in the initial feature data as the first subset of the initial feature data, and selecting feature data greater than or equal to the last selected ratio in the initial feature data as the second subset of the initial feature data, until when the error between the first result and the second result is less than or equal to the error threshold, the second subset of the initial feature data selected last time is used as valid feature data to obtain a second valid feature data set; when the error between the first result and the second result is less than or equal to the error threshold, removing the second subset of the initial feature data from the initial feature data set to obtain a new data set to update the initial feature data set; The initial feature data set is obtained on the business dimension, and feature data on the target business dimension related to the target business scenario in the business dimension is obtained as effective feature data to obtain a third effective feature data set based on the target business scenario, wherein the target business scenario includes repeated hospitalization in low-level hospitals, frequent patients with multiple diseases in low-level hospitals, prescription of drugs in groups, and seniority monitoring of outpatients with chronic diseases, and the business dimension includes one or more of a personnel number dimension, a hospital dimension, a consultation number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features; The union of the first valid feature data set, the second valid feature data set, and the third valid feature data set is taken as the target valid feature data set.

2. The method according to claim 1, characterized in that The tree structure model includes a random forest model, a composite tree model xgboost and a decision tree model lightgbm. The tree structure model is used to obtain the importance of each feature data in the initial feature data set, and the feature data whose importance is greater than the importance threshold is used as the valid feature data to obtain the first valid feature data set, including: Using a random forest model to obtain a first importance of each feature data in the initial feature data set, and taking feature data whose first importance is greater than an importance threshold as valid feature data to obtain a fourth valid feature data set; Using the composite tree model xgboost to obtain the second importance of each feature data in the initial feature data set, and taking the feature data whose second importance is greater than the importance threshold as valid feature data to obtain a fifth valid feature data set; Using the decision tree model lightgbm, obtain the third importance of each feature data in the initial feature data set, and take the feature data whose third importance is greater than the importance threshold as valid feature data to obtain a sixth valid feature data set; The intersection of the fourth valid feature data set, the fifth valid feature data set and the sixth valid feature data set is determined as the first valid feature data set, or the union of the fourth valid feature data set, the fifth valid feature data set and the sixth valid feature data set is determined as the first valid feature data set.

3. The method according to claim 1 or 2, characterized in that The taking the union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set includes: Obtaining an intersection of the first valid feature data set, the second valid feature data set, and the third valid feature data set; The intersection is deduplicated to obtain the target valid feature data set.

4. The method according to claim 3, characterized in that The acquiring, based on the target business scenario, feature data on the target business dimension related to the target business scenario in the business dimension as valid feature data to obtain a third valid feature data set includes: In the case where the target business scenario includes hospitalization in low-level hospitals, a third valid feature data set is obtained based on the feature data on the maximum annual hospitalization cost in the hospital dimension, the number of non-tertiary hospital hospitalization institutions in the department dimension, the non-tertiary hospitals in the department dimension, the number of tertiary hospital hospitalizations in the medical number dimension, and the number of times the hospitalization cost in the hospital dimension is less than the average hospitalization cost of the insured persons in the city as valid feature data; In the case where the target business scenario includes frequent patients with multiple diseases in the low-level hospital, the feature data on the number of hospitalizations in the primary hospital in the medical number dimension, the number of patient diseases in the personnel number dimension, the ratio of patient hospitalization to outpatient care in the insured unit dimension, the total hospitalization expenses of patients in the multi-dimensional features, the average number of items per hospitalization in the multi-dimensional features, the number of primary hospitals visited in the multi-dimensional features, the use of auxiliary drugs in the multi-dimensional features, and the distance from the primary hospital to the hospital in the multi-dimensional features are used as valid feature data to obtain a third valid feature data set; In the case where the target business scenario includes prescribing drugs in groups, a third valid feature data set is obtained based on whether there is a suspected violation of prescribing drugs in groups in the multi-dimensional features, the number of prescribing drugs in groups in the past year in the hospital dimension, the average interval between each violation in the physician dimension, and feature data on the medical institution data in the multi-dimensional features as valid feature data; In the case where the target business scenario includes outpatient chronic disease qualification monitoring, a third valid feature data set is obtained based on the feature data of the long distance to the first-level inpatient hospital, the necessary medication costs in the medical number dimension, the number of necessary medication pick-up times in the medical number dimension, the time interval for medication pick-up in the medical number dimension, the total medication costs in the medical number dimension, the total number of medication pick-up times in the medical number dimension, and other medical types in the multi-dimensional features as valid feature data.

5. The method according to claim 4, characterized in that After performing deduplication processing on the intersection to obtain the target valid feature data set, the method further includes: The target effective feature set is subjected to derivative feature processing to obtain a derived effective feature data set, and the derived effective feature data set is merged into the target effective feature data set to obtain an updated target effective feature data set.

6. A feature validity evaluation device, characterized in that: include: A first acquisition module, used to obtain the importance of each feature data in the initial feature data set by using a tree structure model, and take the feature data whose importance is greater than an importance threshold as valid feature data to obtain a first valid feature data set; A second acquisition module is used to input the initial feature data set into the deep learning neural network model to obtain a first result, obtain a first subset of initial feature data and a second subset of initial feature data based on the initial feature data set, and input the first subset of initial feature data into the deep learning neural network model to obtain a second result, and when the error between the first result and the second result is greater than an error threshold, use the second subset of initial feature data as valid feature data to obtain a second valid feature data set, wherein the initial feature data set includes the first subset of initial feature data and the second subset of initial feature data; A third acquisition module is used to acquire the initial feature data set in the business dimension, and acquire the feature data in the target business dimension related to the target business scenario in the business dimension as effective feature data to obtain a third effective feature data set based on the target business scenario, wherein the target business scenario includes hospitalization in low-level hospitals, frequent patients with multiple diseases in low-level hospitals, prescription of medicines in groups, and outpatient chronic disease seniority monitoring, and the business dimension includes one or more of a personnel number dimension, a hospital dimension, a consultation number dimension, a department dimension, a physician dimension, an insurance unit dimension, and multi-dimensional features; a processing module, configured to take a union of the first valid feature data set, the second valid feature data set, and the third valid feature data set as the target valid feature data set; The deep learning neural network model includes three hidden layers, and the second acquisition module includes: A first input module, used to input the initial feature data set into a deep learning neural network model to obtain a first result; A first selection module, for the deep learning neural network model including three hidden layers; selecting three quarters of the initial feature data set as a first subset of initial feature data, and selecting one quarter of the initial feature data set as a second subset of initial feature data; A second input module, used for inputting the first subset of the initial feature data into the deep learning neural network model to obtain a second result; A second selection module is used for reselecting feature data less than the last selected ratio from the initial feature data set as the first subset of the initial feature data, and reselecting feature data greater than or equal to the last selected ratio from the initial feature data set as the second subset of the initial feature data when the error between the first result and the second result is greater than an error threshold; A third input module, used to re-input the first subset of the initial feature data into the deep learning neural network model to obtain a new first result; a second obtaining module, configured to, when the error between the first result and the second result is greater than an error threshold, repeatedly select feature data less than a proportion selected last time from the initial feature data as the first subset of the initial feature data, and select feature data greater than or equal to a proportion selected last time from the initial feature data as the second subset of the initial feature data, until, when the error between the first result and the second result is less than or equal to the error threshold, obtain a second valid feature data set by using the second subset of the initial feature data selected last time as valid feature data; An updating module is used to remove the second subset of the initial feature data from the initial feature data set to obtain a new data set to update the initial feature data set when the error between the first result and the second result is less than or equal to an error threshold.

7. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the feature effectiveness evaluation method as described in any one of claims 1 to 5 is implemented.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the feature effectiveness evaluation method according to any one of claims 1 to 5 is implemented.

Images