Risk determination method and device for business data, equipment and storage medium
By screening specific insurance policies in group insurance policies and using LightGBM model for feature extraction and training, the problems of low accuracy and insufficient coverage in group insurance data risk prediction are solved, and higher prediction accuracy and efficiency are achieved.
Patent Information
- Application Number
- CN202411136196.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-16
- Publication Date
- 2025-01-03
AI Technical Summary
In the risk prediction of group insurance policy data, the accuracy rate is low and the coverage is insufficient, and the logistic regression model is difficult to fit nonlinear problems. The XGBoost model is inefficient in training and has large memory consumption, so it is impossible to effectively handle category-type features.
By screening policies that meet the policy risk conditions and model debugging conditions in the group policy to be selected, the LightGBM model is used for feature extraction and training, the risk prediction parameters are adjusted, and the fit and prediction accuracy of the model are improved.
Improve the accuracy and efficiency of group policy fraud risk prediction. The LightGBM model is better than logistic regression and XGBoost in data fitting and training speed, and has higher accuracy and coverage.
Smart Images
Figure CN119250979A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of computer technology, and in particular to a method, device, equipment and storage medium for determining risk of business data. Background Art
[0002] At present, the data risk prediction of group insurance policies is mainly based on the risk control rules set by experts. The risk control rules only capture cases under limited conditions, which makes it difficult to control the accuracy, and the generalization is weak, and the adaptability to new data is low. At the same time, the limited rules and conditions cover few risk scenarios, which is more likely to lead to risk penetration.
[0003] In order to improve the accuracy and coverage of data risk prediction, some solutions currently build risk scorecards for claims cases based on logistic regression models or XGBoost models. However, logistic regression models are essentially linear regressions and cannot solve nonlinear problems. There are many scenarios in which insurance policies may encounter risks, and data distributions vary greatly. It is difficult for linear relationships to fit data distributions. In addition, data in group insurance policies also have complex features, outliers, and data imbalances, which make the output results of logistic regression models poor. The XGBoost model cannot handle categorical features, consumes a lot of model training space, and takes a long time to train, resulting in low data processing efficiency. Summary of the invention
[0004] The present application provides a business data risk determination method, apparatus, device and storage medium, aiming to improve the accuracy of determining business data risks and improve the data processing efficiency in the process of business data risk determination.
[0005] In a first aspect, the present application provides a method for determining risk of business data, the method for determining risk of business data comprising the following steps:
[0006] Acquiring business data, wherein the business data includes a plurality of group insurance policies to be selected;
[0007] Determining a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and determining a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected;
[0008] Based on the feature extraction model, feature extraction processing is performed on the first target group insurance policy to obtain training data;
[0009] According to the training data, the preset LightGBM model is trained to obtain a policy risk prediction model;
[0010] adjusting the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model;
[0011] Based on the target insurance policy risk prediction model, the fraud risk value of the group insurance policy to be predicted is determined according to the characteristic data of the group insurance policy to be predicted.
[0012] In a second aspect, the present application further provides a business data risk determination device, the business data risk determination device comprising:
[0013] The insurance policy acquisition module is used to acquire business data, wherein the business data includes a plurality of group insurance policies to be selected;
[0014] An insurance policy screening module, used to determine a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and to determine a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected;
[0015] A data determination module, configured to perform feature extraction processing on the first target group insurance policy based on a feature extraction model to obtain training data;
[0016] A model training module is used to train the preset LightGBM model according to the training data to obtain a policy risk prediction model;
[0017] A parameter adjustment module, used for adjusting the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model;
[0018] The risk prediction module is used to determine the fraud risk value of the group insurance policy to be predicted based on the target insurance policy risk prediction model and the characteristic data of the group insurance policy to be predicted.
[0019] In a third aspect, the present application also provides a computer device, comprising a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the risk determination method for business data as described above are implemented.
[0020] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the steps of the risk determination method for business data as described above are implemented.
[0021] The present application provides a risk determination method, device, equipment and storage medium for business data. The present application screens data by determining a first target group policy that meets the policy risk conditions and a second target group policy that meets the model debugging conditions among the candidate group policies, so as to solve the problems of data imbalance and low data coverage; then the preset LightGBM model is trained with the data corresponding to the first target group policy to obtain a policy risk prediction model, wherein the LightGBM model has better data fitting than the logistic regression model, and has faster training speed and less memory consumption during the model training process compared with the XGBoost model, and the results output by the trained model have higher accuracy and data coverage; then the data corresponding to the second target group policy are used to adjust the risk prediction parameters used to predict the fraud risk value of the group policy in the policy risk prediction model, so as to further improve the prediction accuracy of the policy risk prediction model, and finally, the target policy risk prediction model obtained after the parameter adjustment is used to predict the fraud risk of the group policy to be predicted, thereby improving the efficiency and accuracy of the fraud risk prediction of the group policy. BRIEF DESCRIPTION OF THE DRAWINGS
[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.
[0023] Figure 1 A flowchart of a method for determining risk of business data provided in one embodiment of the present application;
[0024] Figure 2 A schematic diagram of the sub-step flow chart of a method for determining risk of business data provided in an embodiment of the present application;
[0025] Figure 3 A schematic diagram of the sub-step flow chart of a method for determining risk of business data provided in an embodiment of the present application;
[0026] Figure 4 A schematic block diagram of a device for determining risk of business data provided in an embodiment of the present application;
[0027] Figure 5 The present invention is a block diagram showing the structure of a computer device according to an embodiment of the present application. DETAILED DESCRIPTION
[0028] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0029] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.
[0030] The embodiments of the present application provide a method, device, computer equipment and computer-readable storage medium for determining the risk of business data. The method for determining the risk of business data can be applied to a terminal device, which can be an electronic device such as a tablet computer, a laptop computer, a desktop computer, etc. It can also be applied to a server, which can be a separate server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.
[0031] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.
[0032] Please refer to Figure 1 , Figure 1 A flowchart of a method for determining risk of business data provided in one embodiment of the present application.
[0033] like Figure 1 As shown, the business data risk determination method includes steps S101 to S106.
[0034] Step S101: Acquire business data, where the business data includes a plurality of group insurance policies to be selected.
[0035] Exemplarily, business data corresponding to insurance business is obtained from one or more policy data sources to obtain multiple group policies, wherein the group policy is used to indicate the insurance policy generated by the customer insuring in the form of a group. There are significant differences between the data in the group policy and the individual policy. Therefore, the model used to predict the fraud risk of the group policy is different from the model used to predict the fraud risk of the individual policy. If the model for predicting the fraud risk of the individual policy is directly used to predict the group policy, there will be problems such as low data coverage and low prediction accuracy. Therefore, the model for predicting the fraud risk of the group policy is obtained by training to predict the fraud risk value of the group policy, which can improve the accuracy of the group policy fraud risk prediction.
[0036] Step S102: determining a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and determining a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected.
[0037] For example, due to the small amount and high frequency of group insurance claims, the proportion of risk cases is less than 0.1%, and there is an extreme imbalance between positive and negative samples. Therefore, if the model is trained based on the original data distribution, it is impossible to obtain a good model training result, resulting in a low accuracy rate of the risk prediction results of the trained model. By screening the obtained group insurance policies to solve the problem of data imbalance, the training effect of the model can be improved.
[0038] Specifically, by setting different screening conditions, the first target group insurance policies for training the preset model and the second target group insurance policies for further adjusting the model after training are screened out from the acquired group insurance policies, so as to improve the prediction accuracy of the final model.
[0039] In some embodiments, the determining of the first target group policy that meets the policy risk conditions includes: determining the policy risk conditions including that the case closing time of the group policy is within a first time interval, and that the expected claim amount of the group policy is greater than or equal to the premium, the historical accident data status of the group policy is abnormal, the historical inspection data status of the group policy is abnormal, and the group policy does not meet at least one of the insurance claim rejection conditions as the first target group policy.
[0040] In the specific implementation process, the closing time is used to indicate the completion time of the group insurance policy. Specifically, the completion of the group insurance policy includes but is not limited to the completion of the group insurance policy claim, the expiration of the group insurance policy due to the expiration of the insurance period of the group insurance policy, etc. Among them, the time interval length of the first time interval is 2 months to 1 year, and the time when the model needs to be trained is used as the starting time. That is, when model training is required, group insurance policies closed within 2 months from the current time can be obtained, and group insurance policies closed within 1 year from the current time, or group insurance policies closed within 5 months from the current time can be obtained. The acquisition interval is within 2 months to 1 year to improve the timeliness of the data, thereby improving the training effect of the model.
[0041] In the screening of group insurance policies this time, group insurance policies with extremely low risks can be excluded first. Among them, risky group insurance policies include but are not limited to group insurance policies whose claim amount is less than the premium of the group insurance policy, the historical insurance information of the insured in the group insurance policy has been investigated and the information is normal, the insured in the group insurance policy has successfully been insured many times and the historical inspection data is normal, and the insured product is a branded product, etc. In addition, insurance claims rejection conditions include but are not limited to exemption clauses in the policy, claims during the non-insurance period, and the amount does not exceed the free quota, etc. After deleting the group insurance policies with extremely low risks that meet the above conditions from the obtained group insurance policies, the ratio of positive and negative samples can be increased, thereby solving the problem of imbalanced training data for the model.
[0042] In some embodiments, the second target group insurance policy that meets the model debugging conditions includes: determining the group insurance policy whose policy closing time is in a second time interval as the second target group insurance policy, wherein the time length of the second time interval is less than the time length of the first time interval.
[0043] Exemplarily, the second target group policy is used to debug the trained model to determine the training effect of the test model, or to further adjust the parameters of the trained model to improve the accuracy of the model's prediction results.
[0044] During the specific implementation process, the time length corresponding to the second time interval is smaller than the time length of the first time interval. Specifically, if the time length of the first time interval is 1 year, the time length corresponding to the second time interval is 2 months, and the starting time of the second time interval is the current time. That is, the closed insurance policies from the current time to 2 months from the current time are used as the second target insurance policies for debugging the model, thereby improving the accuracy of the model prediction results and the scenario coverage.
[0045] Step S103: Based on the feature extraction model, feature extraction processing is performed on the first target group insurance policy to obtain training data.
[0046] Exemplarily, the first target group insurance policy is input into the feature extraction model so that the feature extraction model extracts corresponding feature data from the first target group insurance policy to obtain training data.
[0047] In the specific implementation process, the feature extraction model includes but is not limited to a feature extraction model composed of a convolutional neural network (CNN) and / or a long short-term memory network (LSTM), so as to extract data from the first target group insurance policy according to the set rules to obtain training data, so that the model can be trained through the training data, thereby completing the construction of the insurance policy risk prediction model.
[0048] See also Figure 2 , Figure 2 A schematic diagram of the sub-step flow chart of a method for determining risk of business data provided in one embodiment of the present application.
[0049] In some embodiments, based on the feature extraction model, feature extraction processing is performed on the first target group insurance policy to obtain training data, including: step S1031, step S1034 to step S1036.
[0050] Step S1031, extracting the policy feature types from the plurality of first target group policies to obtain a plurality of first policy feature types.
[0051] Step S1034: determine the feature missing rate of each of the first insurance policy feature types.
[0052] Step S1035: The first insurance policy feature category whose feature missing rate is less than or equal to the preset feature missing rate threshold is used as the target feature category.
[0053] Step S1036: extract feature data corresponding to the target feature type from each of the first target group insurance policies, and determine the feature data as the training data.
[0054] Exemplarily, feature types include but are not limited to keywords corresponding to data types and / or data items. Specifically, for example, keywords such as policy underwriting model, number of insured units, actual number of insured persons, claim information, current medical information, etc. are extracted from the first target group policy, and the above different keywords are divided into different feature types. Several feature types are extracted from each first target group policy to obtain multiple first policy feature types, and the feature missing rate of each first policy feature type is determined according to the number of times each first policy feature type appears in different first target group policies, thereby determining the first policy feature type for model training according to the feature missing rate.
[0055] In the specific implementation process, for example, there are 100 first target group insurance policies, and among these 100 first target group insurance policies, there are 38 first target group insurance policies containing accident information, then the feature missing rate of accident information is determined to be 62%, and if among these 100 first target group insurance policies, there are 95 first target group insurance policies containing the number of insured units, then the feature missing rate of the number of insured units is determined to be 5%, and when the preset feature missing rate threshold is 60%, the accident information is eliminated, and the number of insured units is determined as the target feature type, so that when training the model, the data corresponding to the number of insured units can be used to train the model.
[0056] It should be noted that the specific keywords extracted above and the corresponding specific feature missing rates are all examples. Different keywords and different feature missing rates can be obtained in different first target group policies. In actual application, the extracted keywords can be set according to needs to obtain the corresponding first policy feature types and calculate the feature missing rates of the first policy feature types. This application does not limit the above keywords.
[0057] See also Figure 3 , Figure 3 A schematic diagram of a sub-step flow chart of a method for determining risk of business data provided in another embodiment of the present application
[0058] In some embodiments, the feature extraction model is based on which feature extraction processing is performed on the first target group insurance policy to obtain training data, including: steps S1031 to S1033 and step S1036.
[0059] Step S1031, extracting the policy feature types from the plurality of first target group policies to obtain a plurality of first policy feature types.
[0060] Step S1032: determine the feature similarity between the first insurance policy feature types.
[0061] Step S1033, screening and merging the first insurance policy feature types whose feature similarity is greater than or equal to the feature similarity threshold to obtain the target feature type.
[0062] Step S1036: extract feature data corresponding to the target feature type from each of the first target group insurance policies, and determine the feature data as the training data.
[0063] Exemplarily, multiple first policy feature types are extracted from multiple first target group policies, and the feature similarity between the multiple first policy feature types is determined to perform feature merging processing. Specifically, data items with different expressions but the same substantive content may appear in multiple first target group policies. For example, the number of insured units is extracted in a first target group policy, and the total number of insured units is extracted in another first target group policy. It should be understood that since the extracted keywords are not the same, they may be divided into two first policy feature types, but the indicated contents are the same. Therefore, by determining the feature similarity between the multiple first policy feature types, the two first policy feature types are merged, thereby avoiding the use of too many and too similar first policy feature types during the training process, improving the training efficiency of the model, and shortening the time of model training.
[0064] In the specific implementation process, the similarity of the keywords corresponding to different first policy feature types is determined, and the feature similarity between the first policy feature types is determined. For example, taking the first policy feature type for the total number of insured units and the first policy feature type for the number of insured units as examples, the feature similarity between the two first policy feature types calculated based on keyword similarity is 85%, and if the feature similarity threshold is 80%, the above two first policy feature types are merged to obtain the target feature type corresponding to the number of insured units, wherein the feature data corresponding to the target feature type is included in the data obtained from the above two first target group policies.
[0065] In other embodiments, after completing the screening and merging of the first policy feature types, the feature missing rate of the merged first policy feature types is calculated, and the first policy feature type whose feature missing rate is less than or equal to the preset feature missing rate threshold is used as the target feature type, thereby completing the screening of the target feature type, that is, determining the training data including steps S1031 to S1036, so as to improve the comprehensiveness and accuracy of the data used for model training, thereby improving the training accuracy of the model.
[0066] In other embodiments, for the first policy feature type whose feature missing rate is less than or equal to the preset feature missing rate threshold, a further IV value calculation is performed, and the first policy feature type whose IV value is less than the preset IV value threshold is determined as the target feature type to achieve further feature type screening. Specifically, the IV value is a statistic used to evaluate the predictive ability of feature variables. In a specific implementation, the IV value of the first policy feature type is determined by weighted summing the WOE (Weight of Evidence) values. Among them, the WOE value is assigned to each first policy feature type, taking into account the proportion of the grouped sample in the overall sample, so that in this way, the IV value can reflect the contribution of the first policy feature type to the prediction result. Exemplarily, the preset IV value threshold is 0.05; it should be understood that the preset feature IV value threshold provided here is an exemplary description, and the present application does not limit the specific preset feature IV value threshold.
[0067] Step S104: train the preset LightGBM model according to the training data to obtain a policy risk prediction model.
[0068] Exemplarily, the training data is input into the preset LightGBM model to train the LightGBM model using the training data. Among them, the LightGBM model belongs to the GBDT series model. The LightGBM model constructs a set of weak learners and combines the output results of multiple weak learners as the final prediction output, where the weak learners include but are not limited to logistic regression, decision trees, etc. The prediction accuracy of the trained LightGBM model is significantly better than the result of a single logistic regression model. Compared with the XGBoost model, which is also a GBDT series model, the LightGBM model uses a histogram algorithm, a gradient-based unilateral sampling algorithm, a level-wise growth strategy, etc. to speed up model training, reduce memory consumption, and reduce model overfitting problems. The model can directly use categorical features, that is, the target feature types in the training data, so that when group insurance policies have data characteristics such as a large number of cases, many types of features, and high importance of categorical features, the LightGBM model has a higher degree of adaptability, so that the trained model has higher prediction accuracy, data coverage, and generalization ability.
[0069] In the specific implementation process, the training data is divided into model training data and model verification data in a ratio of 7:3. It should be understood that the model training data is used to train the preset model, and the model verification data is used to verify the model trained based on the model training data. That is, after the model training is completed using the model training data, the model verification data is input into the trained model. If the output of the trained model meets expectations, it is determined that the model has passed the verification. If the output of the trained model does not meet expectations, it is necessary to continue training the model to ensure the prediction effect of the final output model.
[0070] Step S105: adjust the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model.
[0071] For example, after the preset model is trained using the first target group insurance policy, since the first target group insurance policy covers a longer period of time, the second target group insurance policy with higher timeliness is used as a test set to further adjust the parameters of the trained model, which can improve the model's prediction accuracy.
[0072] In the specific implementation process, the fraud risk values of different group policies are calculated as the test set accuracy and coverage when adjusting the risk prediction parameters. In addition, the accuracy and coverage required to adjust the risk prediction parameters in accordance with actual needs are determined in combination with historical business experience, and then the risk prediction parameters corresponding to the accuracy and coverage are determined.
[0073] Step S106: Based on the target insurance policy risk prediction model, determine the fraud risk value of the group insurance policy to be predicted according to the characteristic data of the group insurance policy to be predicted.
[0074] Exemplarily, after completing the above-mentioned training and parameter adjustment of the model, a target policy risk prediction model is obtained, so as to determine the fraud risk value of the group policy to be predicted based on the target policy risk prediction model and according to the characteristic data in the group policy to be predicted.
[0075] During the specific implementation process, the target policy risk prediction model can extract feature data from the group policy to be predicted, and determine the fraud risk value of the group policy to be predicted based on the extracted feature data, so that the user can determine whether the group policy to be predicted is a fraud case or whether there is a fraud risk based on the output fraud risk value. It should be noted that this application does not limit the specific prediction process of the target policy risk prediction model for the group policy to be predicted.
[0076] In some embodiments, based on the target policy risk prediction model, determining the fraud risk value of the group policy to be predicted according to the characteristic data of the group policy to be predicted includes: determining a plurality of second policy feature types corresponding to the fraud risk value of the group policy to be predicted according to the adjusted risk prediction parameters; grouping the plurality of second policy feature types to obtain a plurality of risk prediction rule sets, each of the risk prediction rule sets including at least two of the second policy feature types; and determining the fraud risk value of the group policy to be predicted according to the risk prediction rule set based on the risk prediction network of the target policy risk prediction model.
[0077] Exemplarily, after completing the adjustment of risk prediction parameters to obtain the target policy risk prediction model, multiple second policy feature types that can be used to predict the fraud risk value of the group policy to be predicted are determined, and a risk prediction rule set is generated based on the second policy feature types, wherein each risk prediction rule set includes at least two of the second policy feature types, so as to predict the fraud risk value of the group policy to be predicted based on the risk prediction rule set, thereby improving the accuracy of the prediction.
[0078] In some embodiments, a network is determined based on the rule set of the target policy risk prediction model, and at least one risk prediction rule set that matches the group policy to be predicted is determined according to the characteristic data of the group policy to be predicted, so as to determine the fraud risk value of the group policy to be predicted in the risk prediction network according to the matching risk prediction rule set.
[0079] Exemplarily, after extracting the characteristic data of the group insurance policy to be predicted, at least one risk prediction rule set that matches the group insurance policy to be predicted is determined, so as to determine the fraud risk value of the group insurance policy to be predicted based on the matched risk prediction rule set. Specifically, after extracting the characteristic data of the group insurance policy to be predicted, the weight value of each characteristic data in the group insurance policy to be predicted is determined, and the characteristic data with a weight value greater than or equal to the weight threshold is matched with the risk prediction rule set, and the risk prediction rule set with a matching value greater than or equal to the preset rule matching threshold is determined as the risk prediction rule set that matches the group insurance policy to be predicted. Among them, the weight threshold is a preset value or is determined according to the average or median of the weight values of each characteristic data, which is not limited in this application. It should be understood that by grouping the characteristic types to obtain the risk prediction rule set, in order to predict the group insurance policy to be predicted, the influence of the association between the characteristic types on the prediction results is enhanced, thereby improving the prediction accuracy of the target insurance policy risk prediction model.
[0080] In other embodiments, after the target policy risk prediction model outputs the fraud risk value of the group policy to be predicted, it also outputs a risk prediction rule set that matches the group policy to be predicted, so that the target policy risk prediction model can provide an explainable description of the abnormal judgment of the group policy to be predicted through the output risk prediction rule set, so as to facilitate business personnel to conduct further review and analysis of group policies with fraud risks and improve the applicability of the model.
[0081] The risk determination method for business data provided in the above embodiment realizes data screening by determining the first target group policy that meets the policy risk conditions and the second target group policy that meets the model debugging conditions among the candidate group policies, so as to solve the problems of data imbalance and low data coverage; then the preset LightGBM model is trained with the data corresponding to the first target group policy to obtain a policy risk prediction model, wherein the LightGBM model has the characteristics of better data fitting, faster model training speed, and less memory consumption; the data corresponding to the second target group policy is then used to adjust the risk prediction parameters used to predict the fraud risk value of the group policy in the policy risk prediction model, thereby further improving the prediction accuracy of the policy risk prediction model; finally, the target policy risk prediction model obtained after the parameter adjustment is used to predict the fraud risk of the group policy to be predicted, thereby improving the efficiency and accuracy of the fraud risk prediction of the group policy.
[0082] See also Figure 4 , Figure 4 1 is a schematic diagram of a business data risk determination device 100 provided in an embodiment of the present application. The business data risk determination device 100 can be configured in a server or a computer device to execute the aforementioned business data risk determination method.
[0083] like Figure 4 As shown, the risk determination device 100 for business data includes: a policy acquisition module 110, a policy screening module 120, a data determination module 130, a model training module 140, a parameter adjustment module 150, and a risk prediction module 160.
[0084] The insurance policy acquisition module 110 is used to acquire business data, wherein the business data includes a plurality of group insurance policies to be selected.
[0085] The insurance policy screening module 120 is used to determine the first target group insurance policies that meet the insurance policy risk conditions among the group insurance policies to be selected, and to determine the second target group insurance policies that meet the model debugging conditions among the group insurance policies to be selected.
[0086] The data determination module 130 is used to perform feature extraction processing on the first target group insurance policy based on the feature extraction model to obtain training data.
[0087] The model training module 140 is used to train the preset LightGBM model according to the training data to obtain a policy risk prediction model.
[0088] The parameter adjustment module 150 is used to adjust the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model.
[0089] The risk prediction module 160 is used to determine the fraud risk value of the group insurance policy to be predicted based on the target insurance policy risk prediction model and according to the characteristic data of the group insurance policy to be predicted.
[0090] Exemplarily, the insurance policy screening module 120 is also used to determine as the first target group insurance policy a group insurance policy that meets at least one of the following policy risk conditions: the case closing time of the group insurance policy is within the first time interval, and the expected claim amount of the group insurance policy is greater than or equal to the premium, the historical accident data status of the group insurance policy is abnormal, the historical inspection data status of the group insurance policy is abnormal, and the group insurance policy does not meet the insurance claim rejection conditions.
[0091] Exemplarily, the insurance policy screening module 120 is further used to determine the group insurance policies whose policy settlement time is in a second time interval as the second target group insurance policies, wherein the time length of the second time interval is shorter than the time length of the first time interval.
[0092] Exemplarily, the data determination module 130 includes a first insurance policy feature extraction submodule, a feature missing rate calculation submodule, a first feature determination submodule, and a training data determination submodule.
[0093] The first insurance policy feature extraction submodule is used to extract insurance policy feature types from multiple insurance policies of the first target group to obtain multiple first insurance policy feature types.
[0094] The feature missing rate calculation submodule is used to determine the feature missing rate of each of the first insurance policy feature types.
[0095] The first feature determination submodule is used to take the first insurance policy feature category whose feature missing rate is less than or equal to a preset feature missing rate threshold as the target feature category.
[0096] The training data determination submodule is used to extract the characteristic data corresponding to the target characteristic type in each of the first target group insurance policies, and determine the characteristic data as the training data.
[0097] Exemplarily, the data determination module 130 further includes a feature similarity calculation submodule.
[0098] The feature similarity calculation submodule is used to determine the feature similarity between the first insurance policy feature types.
[0099] The first feature determination submodule is further used to screen and merge the first insurance policy feature types whose feature similarity is greater than or equal to the feature similarity threshold to obtain the target feature type.
[0100] Exemplarily, the risk prediction module 160 includes a second feature determination submodule, a prediction rule set generation submodule, and a fraud risk value determination submodule.
[0101] The second feature determination submodule is used to sort the multiple prediction frame images according to the order of image generation time sequence, so as to generate a target video according to the sorted prediction frame image sequence.
[0102] The prediction rule set generation submodule is used to group multiple types of the second insurance policy characteristics to obtain multiple risk prediction rule sets, each of which includes at least two of the second insurance policy characteristics types.
[0103] The fraud risk value determination submodule is used to determine the fraud risk value of the group policy to be predicted based on the risk prediction network of the target policy risk prediction model and according to the risk prediction rule set.
[0104] Exemplarily, the risk prediction module 160 further includes a prediction rule set matching submodule.
[0105] The prediction rule set matching submodule is used to determine the network based on the rule set of the target insurance policy risk prediction model, and determine at least one risk prediction rule set that matches the group insurance policy to be predicted according to the characteristic data of the group insurance policy to be predicted.
[0106] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and modules and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.
[0107] The method of the present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld devices or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronic devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.
[0108] Exemplarily, the above method and apparatus may be implemented in the form of a computer program. The computer program may be implemented in Figure 5 Runs on the computer device shown.
[0109] See also Figure 5 , Figure 5 A schematic block diagram of the structure of a computer device provided in an embodiment of the present application. The computer device may be a server or a terminal.
[0110] like Figure 5 As shown, the computer device includes a processor, a memory, and a network interface connected via a system bus, wherein the memory may include a storage medium and an internal memory.
[0111] The storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any one of the risk determination methods for business data.
[0112] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.
[0113] The internal memory provides an environment for the operation of the computer program in the storage medium. When the computer program is executed by the processor, the processor can execute any risk determination method for business data.
[0114] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will appreciate that Figure 5The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.
[0115] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.
[0116] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:
[0117] Acquiring business data, wherein the business data includes a plurality of group insurance policies to be selected;
[0118] Determining a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and determining a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected;
[0119] Based on the feature extraction model, feature extraction processing is performed on the first target group insurance policy to obtain training data;
[0120] According to the training data, the preset LightGBM model is trained to obtain a policy risk prediction model;
[0121] adjusting the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model;
[0122] Based on the target insurance policy risk prediction model, the fraud risk value of the group insurance policy to be predicted is determined according to the characteristic data of the group insurance policy to be predicted.
[0123] In one embodiment, the processor, when determining the first target group insurance policy that meets the insurance policy risk condition, is used to implement:
[0124] The group policy whose policy risk conditions include that the closing time of the group policy is within the first time period, and that the expected claim amount of the group policy is greater than or equal to the premium, the historical accident data status of the group policy is abnormal, the historical inspection data status of the group policy is abnormal, and the group policy does not meet at least one of the insurance claim rejection conditions is determined as the first target group policy.
[0125] In one embodiment, when determining the second target group insurance policy that meets the model debugging condition, the processor is used to implement:
[0126] The group insurance policy with the policy settlement time in the second time interval is determined as the second target group insurance policy, wherein the time length of the second time interval is shorter than the time length of the first time interval.
[0127] In one embodiment, when the processor performs feature extraction processing on the first target group insurance policy based on the feature extraction model to obtain training data, it is used to implement:
[0128] Extracting policy feature types from the plurality of first target group insurance policies to obtain a plurality of first policy feature types;
[0129] Determine the feature missing rate of each of the first policy feature categories;
[0130] The first insurance policy feature category whose feature missing rate is less than or equal to the preset feature missing rate threshold is used as the target feature category;
[0131] Feature data corresponding to the target feature type is extracted from each of the first target group insurance policies, and the feature data is determined as the training data.
[0132] In one embodiment, when the processor performs feature extraction processing on the first target group insurance policy based on the feature extraction model to obtain training data, it is used to implement:
[0133] Extracting policy feature types from the plurality of first target group insurance policies to obtain a plurality of first policy feature types;
[0134] Determining feature similarities between each of the first insurance policy feature types;
[0135] Screening and merging the first policy feature types whose feature similarity is greater than or equal to the feature similarity threshold to obtain a target feature type;
[0136] Feature data corresponding to the target feature type is extracted from each of the first target group insurance policies, and the feature data is determined as the training data.
[0137] In one embodiment, when the processor determines the fraud risk value of the group insurance policy to be predicted based on the target insurance policy risk prediction model and according to the characteristic data of the group insurance policy to be predicted, it is used to implement:
[0138] Determining, based on the adjusted risk prediction parameters, a plurality of second insurance policy feature types corresponding to the fraud risk value of the group insurance policy to be predicted;
[0139] performing grouping processing on a plurality of the second insurance policy feature types to obtain a plurality of risk prediction rule sets, each of the risk prediction rule sets including at least two of the second insurance policy feature types;
[0140] Based on the risk prediction network of the target policy risk prediction model, the fraud risk value of the group policy to be predicted is determined according to the risk prediction rule set.
[0141] In one embodiment, when implementing the risk determination method for business data, the processor is used to implement:
[0142] A network is determined based on the rule set of the target policy risk prediction model, and at least one risk prediction rule set that matches the group policy to be predicted is determined according to the characteristic data of the group policy to be predicted, so as to determine the fraud risk value of the group policy to be predicted in the risk prediction network according to the matching risk prediction rule set.
[0143] It should be noted that technical personnel in the relevant field can clearly understand that, for the convenience and conciseness of description, the specific working process of the computer device described above can refer to the corresponding process in the aforementioned business data risk determination method embodiment, and will not be repeated here.
[0144] An embodiment of the present application also provides a computer-readable storage medium, on which a computer program is stored. The computer program includes program instructions. The method implemented when the program instructions are executed can refer to the various embodiments of the risk determination method for business data of the present application.
[0145] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (SmartMedia Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc., equipped on the computer device.
[0146] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.
[0147] It should also be understood that the term "and / or" used in the specification of this application and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, including these combinations. It should be noted that, in this article, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or system including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or system. In the absence of further restrictions, an element defined by the sentence "including a..." does not exclude the presence of other identical elements in the process, method, article or system including the element.
[0148] The serial numbers of the embodiments of the present application are for description only and do not represent the advantages and disadvantages of the embodiments. The above description is only a specific implementation mode of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.
Claims
1. A method for determining risk of business data, characterized in that: include: Acquiring business data, wherein the business data includes a plurality of group insurance policies to be selected; Determining a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and determining a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected; Based on the feature extraction model, feature extraction processing is performed on the first target group insurance policy to obtain training data; According to the training data, the preset LightGBM model is trained to obtain a policy risk prediction model; adjusting the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model; Based on the target insurance policy risk prediction model, the fraud risk value of the group insurance policy to be predicted is determined according to the characteristic data of the group insurance policy to be predicted.
2. The method for determining the risk of business data according to claim 1, characterized in that: The first target group insurance policy that meets the insurance policy risk conditions is determined, including: The group policy whose policy risk conditions include that the closing time of the group policy is within the first time period, and that the expected claim amount of the group policy is greater than or equal to the premium, the historical accident data status of the group policy is abnormal, the historical inspection data status of the group policy is abnormal, and the group policy does not meet at least one of the insurance claim rejection conditions is determined as the first target group policy.
3. The method for determining the risk of business data according to claim 2, characterized in that: The determining of the second target group insurance policies that meet the model debugging conditions includes: The group insurance policy with the policy settlement time in the second time interval is determined as the second target group insurance policy, wherein the time length of the second time interval is shorter than the time length of the first time interval.
4. The method for determining the risk of business data according to any one of claims 1 to 3, characterized in that: The feature extraction model is used to extract features from the first target group insurance policy to obtain training data, including: Extracting policy feature types from the plurality of first target group insurance policies to obtain a plurality of first policy feature types; Determine the feature missing rate of each of the first policy feature categories; The first insurance policy feature category whose feature missing rate is less than or equal to the preset feature missing rate threshold is used as the target feature category; Feature data corresponding to the target feature type is extracted from each of the first target group insurance policies, and the feature data is determined as the training data.
5. The method for determining the risk of business data according to any one of claims 1 to 3, characterized in that: The feature extraction model is used to extract features from the first target group insurance policy to obtain training data, including: Extracting policy feature types from the plurality of first target group insurance policies to obtain a plurality of first policy feature types; Determining feature similarities between each of the first insurance policy feature types; Screening and merging the first policy feature types whose feature similarity is greater than or equal to the feature similarity threshold to obtain a target feature type; Feature data corresponding to the target feature type is extracted from each of the first target group insurance policies, and the feature data is determined as the training data.
6. The method for determining the risk of business data according to any one of claims 1 to 3, characterized in that: The step of determining the fraud risk value of the group insurance policy to be predicted based on the target insurance policy risk prediction model and according to the characteristic data of the group insurance policy to be predicted includes: Determining, based on the adjusted risk prediction parameters, a plurality of second insurance policy feature types corresponding to the fraud risk value of the group insurance policy to be predicted; performing grouping processing on a plurality of the second insurance policy feature types to obtain a plurality of risk prediction rule sets, each of the risk prediction rule sets including at least two of the second insurance policy feature types; Based on the risk prediction network of the target policy risk prediction model, the fraud risk value of the group policy to be predicted is determined according to the risk prediction rule set.
7. The method for determining the risk of business data according to claim 6, characterized in that: The method further comprises: A network is determined based on the rule set of the target policy risk prediction model, and at least one risk prediction rule set that matches the group policy to be predicted is determined according to the characteristic data of the group policy to be predicted, so as to determine the fraud risk value of the group policy to be predicted in the risk prediction network according to the matching risk prediction rule set.
8. A risk determination device for business data, characterized in that: The business data risk determination device comprises: The insurance policy acquisition module is used to acquire business data, wherein the business data includes a plurality of group insurance policies to be selected; An insurance policy screening module, used to determine a first target group insurance policy that meets the insurance policy risk condition among the group insurance policies to be selected, and to determine a second target group insurance policy that meets the model debugging condition among the group insurance policies to be selected; A data determination module, configured to perform feature extraction processing on the first target group insurance policy based on a feature extraction model to obtain training data; A model training module is used to train the preset LightGBM model according to the training data to obtain a policy risk prediction model; A parameter adjustment module, used for adjusting the risk prediction parameters of the insurance policy risk prediction model according to the second target group insurance policy to obtain a target insurance policy risk prediction model; The risk prediction module is used to determine the fraud risk value of the group insurance policy to be predicted based on the target insurance policy risk prediction model and the characteristic data of the group insurance policy to be predicted.
9. A computer device, characterized in that: The computer device includes a processor, a memory, and a computer program stored in the memory and executable by the processor, wherein when the computer program is executed by the processor, the steps of the method for determining the risk of business data as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein when the computer program is executed by a processor, the steps of the method for determining risk of business data according to any one of claims 1 to 7 are implemented.