Vehicle insurance claim settlement amount prediction method and device, medium and equipment

Abstract

The invention relates to the technical field of computers. The invention specifically discloses a vehicle insurance claim settlement amount prediction method and device, a medium and equipment. The method comprises the following steps: based on a plurality of local sample claim settlement data, using a predetermined initial prediction model to carry out federated model training in this round, and obtaining a first current model parameter; receiving a second current model parameter broadcasted by each non-target client; performing parameter aggregation processing based on the first current model parameter and each second current model parameter to obtain a current aggregation model parameter; at least based on the current aggregation model parameters, determining whether a preset training stop condition is met or not, if the training stop condition is not met, carrying out next-round federated model training until the preset training stop condition is met, and obtaining a target prediction model based on the current aggregation model parameters corresponding to the training round number; and predicting the claim settlement data of the target vehicle based on the target prediction model to obtain a target claim settlement amount. The accuracy of the prediction result can be improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a method, device, medium and equipment for predicting a motor vehicle insurance claim amount. Background Art

[0002] With the continuous development of economy and technology, cars have become more popular, and the number of car insurance purchases has also increased. By purchasing car insurance, car owners can receive financial compensation or cash compensation when their vehicles have accidents or encounter other problems. Therefore, accurate prediction of the claim amount is crucial. By predicting the claim amount of auto insurance, auto insurance fraud can be effectively avoided and auto insurance loss can be verified, which is beneficial to the cost control of insurance companies.

[0003] However, current predictions of auto insurance claims finance are usually based on speculation based on each insurance company's own historical claims data, resulting in inaccurate prediction results. Summary of the invention

[0004] In view of this, the present invention provides a method, device, medium and equipment for predicting the amount of auto insurance claims, the main purpose of which is to solve the current problem of inaccurate prediction of the amount of auto insurance claims.

[0005] To solve the above problems, this application provides a method for predicting the amount of auto insurance claims, which is applied to a target client and includes:

[0006] Based on a number of local sample claim data, a predetermined initial prediction model is used to perform this round of federated model training to obtain corresponding first current model parameters;

[0007] Receiving second current model parameters broadcasted by each non-target client, based on sample claim data local to the non-target client, and obtained by using the predetermined initial prediction model to perform current round of federated model training;

[0008] Based on the first current model parameters and each of the second current model parameters, parameter aggregation processing is performed to obtain current aggregated model parameters;

[0009] At least based on the current aggregate model parameters, determine whether a predetermined training stop condition is met, and when the training stop condition is not met, perform the next round of federated model training until the predetermined training stop condition is met, and obtain target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0010] Based on the target prediction model, the claim amount of the target vehicle is predicted to obtain the target claim amount.

[0011] Optionally, before this round of federated model training, the method for predicting the amount of auto insurance claims also includes:

[0012] Perform feature extraction on each local sample claim data to obtain an initial feature vector set corresponding to each sample claim data;

[0013] Using a predetermined gradient boosting decision tree model to pre-train each initial feature vector set, to obtain a pre-trained model;

[0014] Based on each initial feature vector set and the pre-trained model, determining the importance of each initial feature vector set;

[0015] The initial feature vector sets are screened based on their importance to obtain a target feature vector set for training the federated model.

[0016] The method for predicting the amount of automobile insurance claims also includes:

[0017] Determine a first feature dimension of a target feature vector set;

[0018] Obtaining a second feature dimension of a non-target feature vector set obtained after each non-target client extracts features from the non-target client's local sample claims data;

[0019] determining whether the second characteristic dimensions and the first characteristic dimensions are consistent with each other, and determining a target characteristic dimension if they are inconsistent;

[0020] Each target feature vector set is padded based on the target feature dimension to obtain a plurality of padded feature vector sets for training the federated model.

[0021] Optionally, before performing parameter aggregation processing, the method for predicting the amount of automobile insurance claims further includes:

[0022] Obtain the amount of local sample claims data of each non-target client for federated model training;

[0023] The performing parameter aggregation processing based on the first current model parameter and each of the second current model parameters to obtain current aggregated model parameters specifically includes:

[0024] Determine a weight corresponding to each non-target client based on each of the data amounts;

[0025] Based on the first current model parameter, each of the second current model parameters and the weight corresponding to each non-target client, parameter aggregation processing is performed to obtain current aggregated model parameters.

[0026] Optionally, determining whether a predetermined training stop condition is met at least based on the current aggregate model parameters specifically includes:

[0027] Calculate the loss value based on the current aggregate model parameters;

[0028] Determine whether a predetermined training stop condition is met based on the loss value.

[0029] Optionally, the method for predicting the amount of automobile insurance claims also includes:

[0030] Determine the current number of training rounds;

[0031] Determine whether a predetermined training stop condition is met based on the current number of training rounds.

[0032] Optionally, before the next round of federated model training, the method for predicting the amount of auto insurance claims also includes:

[0033] Based on the current aggregate model parameters and the first current model parameters corresponding to this round of federated model training, a training gradient is determined so that the first current model parameters are adjusted based on the training gradient when the next round of federated model training is performed.

[0034] In order to solve the above problems, the present application provides a prediction device for the amount of automobile insurance claims, comprising:

[0035] A training module, used to perform current round of federated model training based on a number of local sample claim data using a predetermined initial prediction model to obtain corresponding first current model parameters;

[0036] A receiving module, used for receiving the second current model parameters broadcasted by each non-target client, based on the sample claim data of the non-target client, and obtained by using the predetermined initial prediction model to perform this round of federated model training;

[0037] an aggregation module, configured to perform parameter aggregation processing based on the first current model parameter and each of the second current model parameters to obtain a current aggregated model parameter;

[0038] An acquisition module, configured to determine whether a predetermined training stop condition is satisfied based at least on the current aggregate model parameters, and when the training stop condition is not satisfied, perform the next round of federated model training until the predetermined training stop condition is satisfied, and obtain target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0039] The prediction module is used to predict the claim amount of the target vehicle's claim data based on the target prediction model to obtain a target claim amount.

[0040] To solve the above problems, the present application provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the amount of a motor vehicle insurance claim described in any one of the above items are implemented.

[0041] To solve the above problems, the present application provides an electronic device, comprising at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any of the above-mentioned methods for predicting the amount of a motor vehicle insurance claim when executing the computer program on the memory.

[0042] The method, device, medium and equipment for predicting the amount of auto insurance claims in the present application, by adopting a federated learning approach, aggregates the first current model parameters obtained by local training of the target client and the second current model parameters of each participant (non-target client) to obtain aggregated model parameters, and when the aggregated model parameters meet the training stop conditions, obtains the target model parameters based on the aggregated model parameters, thereby making the final target model parameters more accurate, and realizing data sharing while ensuring the data privacy of each participant, that is, realizing the final target prediction model based on the accurate training of the data of each participant, thereby improving the accuracy of the prediction results.

[0043] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0045] Figure 1 This is a flow chart of a method for predicting a motor vehicle insurance claim amount according to an embodiment of the present application;

[0046] Figure 2 This is a structural block diagram of a device for predicting the amount of a motor vehicle insurance claim according to another embodiment of the present application;

[0047] Figure 3 This is a structural block diagram of an electronic device according to another embodiment of the present application. DETAILED DESCRIPTION

[0048] Various aspects and features of the present application are described herein with reference to the accompanying drawings.

[0049] It should be understood that various modifications may be made to the embodiments of the present application. Therefore, the above description should not be considered as limiting, but only as an example of an embodiment. Other modifications within the scope and spirit of the present application will occur to those skilled in the art.

[0050] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the general description of the present application given above and the detailed description of the embodiments given below, serve to explain the principles of the present application.

[0051] These and other characteristics of the present application will become apparent from the following description of a preferred form of embodiment given as a non-limiting example with reference to the accompanying drawings.

[0052] It should also be understood that although the present application has been described with reference to some specific examples, those skilled in the art will be able to readily implement many other equivalent forms of the present application.

[0053] The above and other aspects, features and advantages of the present application will become more apparent in view of the following detailed description when taken in conjunction with the accompanying drawings.

[0054] Specific embodiments of the present application are described hereinafter with reference to the accompanying drawings; however, it should be understood that the embodiments applied for are merely examples of the present application, which may be implemented in a variety of ways. Well-known and / or repeated functions and structures are not described in detail to avoid unnecessary or redundant details that obscure the present application. Therefore, the specific structural and functional details applied for herein are not intended to be limiting, but merely serve as a basis and representative basis for the claims to teach those skilled in the art to use the present application in a variety of ways with substantially any suitable detailed structure.

[0055] This specification may use the phrases "in one embodiment," "in another embodiment," "in yet another embodiment," or "in other embodiments," all of which may refer to one or more of the same or different embodiments according to the present application.

[0056] The embodiment of the present application provides a method for predicting the amount of auto insurance claims, which can be applied to the scenario of auto insurance claims in the financial field. The specific execution subject can be a terminal or electronic device corresponding to any insurance company (participant) among several insurance company objects (participants) participating in federated learning. Any participant is taken as the target participant, and the terminal corresponding to the target participant is the target terminal, and the terminals corresponding to other participants are non-target terminals, such as Figure 1 As shown, the method in this example includes the following steps:

[0057] Step S101, based on a number of local sample claim data, using a predetermined initial prediction model to perform current round of federated model training to obtain corresponding first current model parameters;

[0058] In this step, the target terminal can use the local historical claims data as sample claims data, and then extract features from the sample claims data to obtain several feature vector sets, so as to perform the first round / current round of federated training on the initial prediction model based on each feature vector set to obtain the corresponding first current model parameters. After obtaining the first current model parameters, the first current parameters can be broadcasted so that each non-target client can obtain the first current model parameters, which is convenient for each non-target client to perform federated model training.

[0059] In this step, by extracting features from each sample claims data, a feature set corresponding to each sample claims data can be obtained, thereby obtaining several feature sets, each of which includes: an initial feature vector set of feature contents such as vehicle information, accident information, maintenance information, and historical maintenance information. Among them, vehicle attributes include: vehicle parameter configuration, vehicle condition information, vehicle monitoring information, driver information, driving behavior information, etc. Accident information includes: accident exploration information, damage assessment agency data, damage assessor information, etc. Maintenance information includes: working hours information, parts information, maintenance pricing, and maintenance personnel information. Historical maintenance information includes: historical claim amounts, etc. After obtaining the feature set, the features in the feature set can be converted into vectors to obtain each feature vector set.

[0060] Step S102, receiving the second current model parameters broadcasted by each non-target client, based on the sample claim data of the non-target client, and obtained by using the predetermined initial prediction model to perform this round of federated model training;

[0061] In this step, while the target client performs model training based on local sample claims data, each non-target client (other participating party) will also perform this round of federated model training based on the local sample claims data of the non-target client and using the predetermined initial prediction model, thereby obtaining the second current model parameters corresponding to each non-target client. Afterwards, each non-target client can broadcast its own second current model parameters, so that any client (participant) can obtain the model parameters of other clients (other participating parties), laying the foundation for subsequent parameter aggregation based on the model parameters of each participating party (client).

[0062] Step S103, performing parameter aggregation processing based on the first current model parameters and each of the second current model parameters to obtain current aggregated model parameters;

[0063] During the specific implementation of this step, the target client may first determine its own weight and the weights of each non-target client, and then perform parameter aggregation processing on the first current model parameter and each second current model parameter according to the weights.

[0064] Step S104, determining whether a predetermined training stop condition is met at least based on the current aggregate model parameters, and when the training stop condition is not met, performing the next round of federated model training until the predetermined training stop condition is met, obtaining target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0065] In this step, the predetermined training stop condition can be a predetermined loss threshold. Specifically, the loss value can be calculated according to the aggregation model parameters using a predetermined loss function, and then the loss value is compared with the loss threshold to determine whether the training stop condition is met. That is, when the calculated loss value is less than the loss threshold, that is, when the loss function converges, it means that the predetermined training stop condition is met. Therefore, the current aggregation model parameters corresponding to the current training round number can be used as the target model parameters to obtain the target prediction model.

[0066] Step S105, predicting the claim amount of the target vehicle's claim data based on the target prediction model to obtain a target claim amount.

[0067] In this step, after the target prediction model is obtained, the target prediction model can be used to predict the claims data that need to be predicted for the claim amount, thereby obtaining the prediction result.

[0068] The method for predicting the amount of auto insurance claims in this embodiment adopts a federated learning approach to aggregate the first current model parameters obtained by local training of the target client and the second current model parameters of each participant (non-target client) to obtain aggregated model parameters, and when the aggregated model parameters meet the training stop conditions, the target model parameters are obtained based on the aggregated model parameters, thereby making the final target model parameters more accurate, and realizing data sharing while ensuring the data privacy of each participant, that is, realizing the final target prediction model based on the accurate training of the data of each participant, thereby improving the accuracy of the prediction results.

[0069] Another embodiment of the present application provides a method for predicting the amount of auto insurance claims. In this embodiment, before the target client performs this round of federated model training, that is, before the first round of federated model training, in order to make the training results more accurate, the feature vector set corresponding to each local sample claim data can be screened to obtain a target feature vector set for federal model training. The specific screening process is: feature extraction is performed on each local sample claim data to obtain an initial feature vector set corresponding to each sample claim data; a predetermined gradient boosting decision tree model is used to perform model pre-training on each initial feature vector set to obtain a pre-trained model; based on each initial feature vector set and the pre-trained model, the importance of each initial feature vector set is determined; based on the importance of each initial feature vector set, each initial feature vector set is screened to obtain a target feature vector set for federal model training.

[0070] Similarly, other non-target clients participating in federated learning will also screen the feature vector set corresponding to the local sample claims data to obtain the target sample training set for model training. The screening process is the same as the above process and will not be repeated here. In this embodiment, by screening the feature vector set, a target feature vector set with a high correlation with the model training can be obtained, thereby laying the foundation for subsequent accurate training to obtain a prediction model.

[0071] Another embodiment of the present application provides a method for predicting the amount of auto insurance claims. In this embodiment, in order to make the dimension of the feature vector used for model training of the target client and each non-target client consistent, after obtaining the target feature vector set, the first feature dimension of the target feature vector set can be further determined. Then, the second feature dimension of the non-target feature vector set obtained after each non-target client extracts the feature of the local sample claim data of the non-target client is obtained; the feature dimension can be broadcasted by broadcasting so that each client can obtain the feature dimension of other clients. Determine whether the second feature dimension and the first feature dimension are consistent. If they are inconsistent, determine the target feature dimension; finally, based on the target feature dimension, each target feature vector set is padded to obtain several padded feature vector sets for federated model training. That is, if the first feature dimension of the target client is inconsistent with the second dimension of each non-target client, the first feature dimension or the second feature dimension with the largest feature dimension can be used as the target feature dimension, and then each target feature vector set is padded based on the target feature dimension. In this embodiment, since any client can obtain the feature dimensions of other clients, the target feature dimensions determined by each client (target client and non-target clients) are the same, that is, the target feature dimensions are the largest feature dimensions among the feature dimensions corresponding to each client. In this embodiment, by performing feature completion processing, that is, using data features as alignment guides, the data of each client (participant) is sample aligned, which can expand the sample space and provide a guarantee for subsequent accurate training to obtain the target prediction model.

[0072] Another embodiment of the present application provides a method for predicting the amount of auto insurance claims. In this embodiment, before performing sample parameter aggregation processing, the target client can also obtain the amount of local sample claims data used by each non-target client for federal model training; and determine the target client's own weight and the weight corresponding to each non-target client based on the amount of data participating in the local model training and the amount of data of each non-target client. Therefore, when performing parameter aggregation processing, parameter aggregation processing can be performed based on the first current model parameters, the weights corresponding to the target clients of each of the second current model parameters, and the weights corresponding to each non-target client to obtain the current aggregated model parameters. In this embodiment, the corresponding weights are determined based on the amount of data participating in the model training of each client (target client and non-target client), so that the determination of the weight value can be more reasonable and accurate.

[0073] For example, the amount of data participating in the model training of the target client a is 1000, the amount of data participating in the model training of the non-target client b is 800, the amount of data participating in the model training of the non-target client c is 1000, and the amount of data participating in the model training of the non-target client d is 600. Then the ratio of the number of target client a to non-target client b, non-target client c and non-target client d is 5:4:5:3. Therefore, the weight of the target client a is 5 / 17, the weight of the non-target client b is 4 / 17, the weight of the non-target client c is 5 / 17, and the weight of the non-target client d is 3 / 17.

[0074] Another embodiment of the present application provides a method for predicting the amount of auto insurance claims. In this embodiment, when it is determined that the predetermined training stop condition is met, it can be further determined in combination with the number of training rounds, that is, when the loss value calculated based on the aggregate model parameters is less than the predetermined loss threshold or the number of training rounds is greater than the predetermined round threshold, it is determined that the training stop condition is met. That is, by calculating whether the loss function converges or whether the number of iterative training rounds reaches the upper limit, it is determined whether to stop model training. This can avoid the problem of continuous model training due to the inability of the loss function to converge.

[0075] Another embodiment of the present application provides a method for predicting the amount of a motor vehicle insurance claim. In this embodiment, when it is determined based on the current aggregate model parameters that the predetermined training stop condition is not met, that is, when the loss value calculated based on the current aggregate model parameters using the loss function is not less than the predetermined loss threshold, and the number of training rounds reaches the predetermined round threshold, the next round of federated model training is required. Thus, the target client can determine the training gradient based on the current aggregate model parameters and the first current model parameters; then adjust the first current model parameters based on the training gradient to perform the next round of federated model training. In this embodiment, by determining the training gradient based on the current aggregate model parameters, the adjustment of the model parameters can be more reasonable in the next round of model training, which lays the foundation for obtaining the final local model through express and accurate training. Subsequently, the target prediction model can be obtained based on the local model obtained through training and the local models of each non-target client.

[0076] Another embodiment of the present application provides a method for predicting the amount of a motor vehicle insurance claim. In this embodiment, in order to ensure data security and prevent the model parameters obtained by each participant (target client and each non-target client) from being stolen, before broadcasting the current model parameters obtained by training, the current model can also be encrypted using a predetermined encryption algorithm. That is, the target client encrypts the first current model parameter using a predetermined encryption algorithm, and then broadcasts the encrypted first current model parameter, and each non-target client encrypts its own second current model parameter using a predetermined encryption algorithm, and then broadcasts the encrypted second current model parameter. Thus, after the target client obtains the encrypted second current model parameters broadcasted by each non-target client, it can be securely aggregated based on homomorphic encryption, that is, using a decryption algorithm corresponding to the encryption algorithm to decrypt each encrypted second current model parameter, and then perform parameter aggregation processing to obtain the aggregated model parameters.

[0077] In order to further explain the above embodiment, the process of each (client) participant in this application performing federated model training is specifically described below. The specific process is as follows:

[0078] Step 1: Each client participating in federated learning constructs an initial feature vector set based on its own sample claims data;

[0079] In this step, each client can construct a feature vector set based on its own sample claims data. Since the compensation for auto insurance damages is not only related to fixed data such as maintenance items, but also includes dynamic data of the vehicle and dynamic data of related personnel. Therefore, the sample claims data can be feature extracted and converted into vectors, thereby obtaining an initial feature vector set corresponding to the sample claims data, which contains feature content such as vehicle information, accident information, maintenance information, and historical maintenance information. Among them, vehicle attributes include: vehicle parameter configuration, vehicle condition information, vehicle monitoring information, driver information, driving behavior information, etc. Accident information includes: accident exploration information, damage assessment agency data, damage assessor information, etc. Maintenance information includes: working hours information, parts information, maintenance pricing, and maintenance personnel information. Historical maintenance information includes: historical claim amounts, etc.

[0080] Step 2: Each client screens its own initial feature vector set to obtain a target feature vector set for federated model training;

[0081] In this step, each client can use a predetermined gradient boosting decision tree model to pre-train the local initial feature vector set to obtain a pre-trained model; based on each initial feature vector set and the pre-trained model, determine the importance of each initial feature vector set; based on the importance of each initial feature vector set, screen each initial feature vector set to obtain a target feature vector set for federal model training.

[0082] Step 3: Each client performs sample alignment based on data features to expand the sample space.

[0083] In this step, each client can broadcast the feature dimensions by broadcasting, so that any client can obtain the feature dimensions of other clients. Then, each client can determine the target feature dimension based on the feature dimensions of the local feature vector and the feature dimensions of other clients, and finally complete the local feature vector set based on the target feature dimension to obtain several completed feature vector sets for federated model training.

[0084] Step 4: Each client participating in federated learning performs this round of model training based on the local completed feature vector set using the pre-agreed initial prediction model to obtain the current model parameters corresponding to each client;

[0085] Step 5: Each client broadcasts the current model parameters obtained in this round of training;

[0086] In this step, each client can share the current model parameters by broadcasting the local current model parameters, so that each client participating in federated learning can obtain the current model parameters of other clients.

[0087] In this step, before broadcasting the current model parameters, each client may also encrypt the local current model parameters based on a pre-agreed encryption algorithm and then broadcast them, thereby ensuring the security of the model data.

[0088] Step 6: Each client obtains the current model parameters broadcast by other clients, performs parameter aggregation based on the local current model parameters and the current model parameters corresponding to other clients, and obtains aggregated model parameters.

[0089] In this step, each client can obtain the amount of data used by other clients for federated model training; then, based on the amount of data participating in the local model training and the amount of data corresponding to other clients, determine its own weight and the weights of other clients; finally, perform a federated weighted average calculation on the current model parameters corresponding to each client according to the weight of each client, thereby obtaining the aggregated model parameters;

[0090] In this step, before performing the federated weighted average calculation, the decryption algorithm corresponding to the encryption algorithm can be used to decrypt the current model parameters encrypted by each client, and then the federated weighted average calculation is performed. Specifically, a secure aggregation based on homomorphic encryption is performed, and the federated weighted average calculation is performed here.

[0091] Step 7: Based on the current aggregate model parameters and the current number of training rounds, determine whether the training stop condition is met; if the training stop condition is met, use the current aggregate model parameters as the target model parameters to obtain the target prediction model; if the training stop condition is not met, execute step 8.

[0092] Step 8: Based on the current aggregate model parameters and the current model parameters obtained by the local training of the client, determine the training gradient for each client to perform the next round of federated model training, and perform the next round of training on the local current prediction model based on the training gradient and the local target feature vector set, re-obtain the current model parameters, and return to step 5;

[0093] In this embodiment, by adopting the above-mentioned training process, the decentralized federated peer-to-peer network computing architecture can be used for model training. There is no central server, and all interactions are conducted directly between the participants (each terminal device). On the basis of ensuring the data privacy of each participant, multiple models are jointly trained to expand the sample space and improve the accuracy of the damage assessment model. All participants can benefit from the global model obtained through training, thereby realizing intelligent damage assessment, making the prediction of the final claim amount more accurate, reducing auto insurance claims expenditures, helping insurance companies control costs, increase revenue and further expand their business.

[0094] Another embodiment of the present application provides a device for predicting the amount of automobile insurance claims, such as Figure 2 As shown, including:

[0095] The training module 11 is used to perform this round of federated model training based on a number of local sample claim data using a predetermined initial prediction model to obtain corresponding first current model parameters;

[0096] A receiving module 12 is used to receive the second current model parameters broadcasted by each non-target client, based on the sample claim data of the non-target client, and obtained by using the predetermined initial prediction model to perform this round of federated model training;

[0097] an aggregation module 13, configured to perform parameter aggregation processing based on the first current model parameter and each of the second current model parameters to obtain a current aggregated model parameter;

[0098] An obtaining module 14 is used to determine whether a predetermined training stop condition is met based on at least the current aggregate model parameters, and when the training stop condition is not met, perform the next round of federated model training until the predetermined training stop condition is met, and obtain target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0099] The prediction module 15 is used to predict the claim amount of the target vehicle's claim data based on the target prediction model to obtain a target claim amount.

[0100] In the specific implementation process of this embodiment, the prediction device of the automobile insurance claim amount further includes: an extraction module and a screening module; the extraction module is used to: perform feature extraction on each local sample claim data to obtain an initial feature vector set corresponding to each sample claim data;

[0101] The screening module is used to: use a predetermined gradient boosting decision tree model to pre-train each initial feature vector set to obtain a pre-trained model; determine the importance of each initial feature vector set based on each initial feature vector set and the pre-trained model; screen each initial feature vector set based on the importance of each initial feature vector set to obtain a target feature vector set for federal model training.

[0102] In the specific implementation process of this embodiment, the device for predicting the amount of auto insurance claims also includes a first determination module, which is used to: determine a first feature dimension of a target feature vector set; obtain a second feature dimension of a non-target feature vector set obtained after each non-target client performs feature extraction on local sample claim data of the non-target client; determine whether the second feature dimensions and the first feature dimensions are consistent, and if they are inconsistent, determine the target feature dimension; and perform padding processing on each target feature vector set based on the target feature dimension to obtain a number of padding feature vector sets for federal model training.

[0103] In the specific implementation process of this embodiment, the prediction device of the automobile insurance claim amount further includes a second determination module, which is used to: before performing parameter aggregation processing, obtain the data volume of local sample claim data of each non-target client for federated model training;

[0104] The aggregation module is specifically used to: determine the weight corresponding to each non-target client based on each data volume; perform parameter aggregation processing based on the first current model parameters, each second current model parameter and the weight corresponding to each non-target client to obtain the current aggregation model parameters.

[0105] In the specific implementation process of this embodiment, the prediction device for the motor insurance claim amount also includes a calculation module, which is used to: calculate the loss value based on the current aggregation model parameters; the acquisition module is used to: determine whether a predetermined training stop condition is met based on the loss value.

[0106] In the specific implementation process of this embodiment, the prediction device for the amount of automobile insurance claims further includes a third determination module, and the third determination module is used to: determine the current number of training rounds;

[0107] The obtaining module is further used to determine whether a predetermined training stop condition is met based on the current number of training rounds.

[0108] In the specific implementation process of this embodiment, the device for predicting the amount of auto insurance claims also includes a fourth determination module, and the fourth determination module is used to: before the next round of federal model training, determine the training gradient based on the current aggregated model parameters corresponding to this round of federal model training and the first current model parameters, so as to adjust the first current model parameters based on the training gradient when the next round of federal model training is performed.

[0109] The device for predicting the amount of auto insurance claims in this embodiment adopts a federated learning method to aggregate the first current model parameters obtained by local training of the target client and the second current model parameters of each participant (non-target client) to obtain aggregated model parameters, and when the aggregated model parameters meet the training stop conditions, the target model parameters are obtained based on the aggregated model parameters, thereby making the final target model parameters more accurate, and realizing data sharing while ensuring the data privacy of each participant, that is, realizing the final target prediction model based on the accurate training of the data of each participant, thereby improving the accuracy of the prediction results.

[0110] Another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, and when the computer program is executed by a processor, the following method steps are implemented:

[0111] Step 1: Based on a number of local sample claim data, use the predetermined initial prediction model to perform this round of federated model training to obtain the corresponding first current model parameters;

[0112] Step 2: receiving the second current model parameters obtained by training the current round of federated model based on the sample claim data broadcasted by each non-target client and the local non-target client using the predetermined initial prediction model;

[0113] Step 3: performing parameter aggregation processing based on the first current model parameters and each of the second current model parameters to obtain current aggregated model parameters;

[0114] Step 4: determining whether a predetermined training stop condition is met at least based on the current aggregate model parameters, and when the training stop condition is not met, performing the next round of federated model training until the predetermined training stop condition is met, obtaining target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0115] Step 5: Predict the claim amount of the target vehicle based on the target prediction model to obtain the target claim amount.

[0116] The specific implementation process of the above method steps can be found in the above-mentioned embodiment of the method for predicting the claim amount of any motor vehicle insurance, and this embodiment will not be repeated here.

[0117] The storage medium in the present application, by adopting a federated learning approach, aggregates the first current model parameters obtained by local training of the target client and the second current model parameters of each participant (non-target client) to obtain aggregated model parameters, and when the aggregated model parameters meet the training stop conditions, obtains the target model parameters based on the aggregated model parameters, thereby making the final target model parameters more accurate, and realizing data sharing while ensuring the data privacy of each participant, that is, realizing the final target prediction model obtained by accurate training based on the data of each participant, thereby improving the accuracy of the prediction results.

[0118] Another embodiment of the present application provides an electronic device, such as Figure 3 As shown, at least a memory 1 and a processor 2 are included, the memory 1 stores a computer program, and the processor 2 implements the following method steps when executing the computer program on the memory 1:

[0119] Step 1: Based on a number of local sample claim data, use the predetermined initial prediction model to perform this round of federated model training to obtain the corresponding first current model parameters;

[0120] Step 2: receiving the second current model parameters obtained by training the current round of federated model based on the sample claim data broadcasted by each non-target client and the local non-target client using the predetermined initial prediction model;

[0121] Step 3: performing parameter aggregation processing based on the first current model parameters and each of the second current model parameters to obtain current aggregated model parameters;

[0122] Step 4: determining whether a predetermined training stop condition is met at least based on the current aggregate model parameters, and when the training stop condition is not met, performing the next round of federated model training until the predetermined training stop condition is met, obtaining target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model;

[0123] Step 5: Predict the claim amount of the target vehicle based on the target prediction model to obtain the target claim amount.

[0124] The specific implementation process of the above method steps can be found in the above-mentioned embodiment of the method for predicting the claim amount of any motor vehicle insurance, and this embodiment will not be repeated here.

[0125] The electronic device in the present application adopts a federated learning method to aggregate the first current model parameters obtained by local training of the target client and the second current model parameters of each participant (non-target client) to obtain aggregated model parameters, and when the aggregated model parameters meet the training stop conditions, the target model parameters are obtained based on the aggregated model parameters, thereby making the final target model parameters more accurate, and realizing data sharing while ensuring the data privacy of each participant, that is, realizing the final target prediction model based on the accurate training of the data of each participant, thereby improving the accuracy of the prediction results.

[0126] The above embodiments are only exemplary embodiments of the present application and are not intended to limit the present application. The protection scope of the present application is defined by the claims. Those skilled in the art may make various modifications or equivalent substitutions to the present application within the essence and protection scope of the present application, and such modifications or equivalent substitutions shall also be deemed to fall within the protection scope of the present application.

Claims

1. A method for predicting the amount of automobile insurance claims, applied to a target client, characterized in that: include: Based on a number of local sample claim data, a predetermined initial prediction model is used to perform this round of federated model training to obtain corresponding first current model parameters; Receiving second current model parameters broadcasted by each non-target client, based on sample claim data local to the non-target client, and obtained by using the predetermined initial prediction model to perform current round of federated model training; Based on the first current model parameters and each of the second current model parameters, parameter aggregation processing is performed to obtain current aggregated model parameters; At least based on the current aggregate model parameters, determine whether a predetermined training stop condition is met, and when the training stop condition is not met, perform the next round of federated model training until the predetermined training stop condition is met, and obtain target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model; Based on the target prediction model, the claim amount of the target vehicle is predicted to obtain the target claim amount.

2. The method according to claim 1, characterized in that Before performing this round of federated model training, the method further includes: Perform feature extraction on each local sample claim data to obtain an initial feature vector set corresponding to each sample claim data; Using a predetermined gradient boosting decision tree model to pre-train each initial feature vector set, to obtain a pre-trained model; Based on each initial feature vector set and the pre-trained model, determining the importance of each initial feature vector set; The initial feature vector sets are screened based on their importance to obtain a target feature vector set for training the federated model.

3. The method according to claim 2, characterized in that The method further comprises: Determine a first feature dimension of a target feature vector set; Obtaining a second feature dimension of a non-target feature vector set obtained after each non-target client extracts features from the non-target client's local sample claims data; determining whether the second characteristic dimensions and the first characteristic dimensions are consistent with each other, and determining a target characteristic dimension if they are inconsistent; Each target feature vector set is padded based on the target feature dimension to obtain a plurality of padded feature vector sets for training the federated model.

4. The method according to claim 1, characterized in that Before performing parameter aggregation processing, the method further includes: Obtain the amount of local sample claims data of each non-target client for federated model training; The performing parameter aggregation processing based on the first current model parameter and each of the second current model parameters to obtain current aggregated model parameters specifically includes: Determine a weight corresponding to each non-target client based on each of the data amounts; Based on the first current model parameter, each of the second current model parameters and the weight corresponding to each non-target client, parameter aggregation processing is performed to obtain current aggregated model parameters.

5. The method according to claim 1, characterized in that The determining whether a predetermined training stop condition is met based at least on the current aggregate model parameters specifically includes: Calculate the loss value based on the current aggregate model parameters; Determine whether a predetermined training stop condition is met based on the loss value.

6. The method according to claim 5, characterized in that The method further comprises: Determine the current number of training rounds; Determine whether a predetermined training stop condition is met based on the current number of training rounds.

7. The method according to claim 1, characterized in that Before performing the next round of federated model training, the method further includes: Based on the current aggregate model parameters and the first current model parameters corresponding to this round of federated model training, a training gradient is determined so that the first current model parameters are adjusted based on the training gradient when the next round of federated model training is performed.

8. A device for predicting the amount of automobile insurance claims, characterized in that: include: A training module, used to perform current round of federated model training based on a number of local sample claim data using a predetermined initial prediction model to obtain corresponding first current model parameters; A receiving module, used for receiving the second current model parameters broadcasted by each non-target client, based on the sample claim data of the non-target client, and obtained by using the predetermined initial prediction model to perform this round of federated model training; an aggregation module, configured to perform parameter aggregation processing based on the first current model parameter and each of the second current model parameters to obtain a current aggregated model parameter; An acquisition module, configured to determine whether a predetermined training stop condition is satisfied based at least on the current aggregate model parameters, and when the training stop condition is not satisfied, perform the next round of federated model training until the predetermined training stop condition is satisfied, and obtain target model parameters based on the current aggregate model parameters corresponding to the number of training rounds to obtain a target prediction model; The prediction module is used to predict the claim amount of the target vehicle's claim data based on the target prediction model to obtain a target claim amount.

9. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method for predicting the amount of a motor vehicle insurance claim as described in any one of claims 1 to 7 are implemented.

10. An electronic device, characterized in that: The method comprises at least a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method for predicting the amount of automobile insurance claim as described in any one of claims 1 to 7 when executing the computer program on the memory.

Images