An intelligent prediction method and system for injury claims based on big data

Abstract

The present invention discloses a method and system for intelligent prediction of injury claims based on big data, which belongs to the technical field of claims prediction. The method includes data preparation, data preprocessing, injury level prediction, injury claims prediction and intelligent prediction report generation. The present invention adopts a convolutional bidirectional long-term and short-term network combined with an attention mechanism to predict the injury level, better process complex data, enhance the robustness and accuracy of the model, effectively assist subsequent injury claims prediction, and improve the reliability and effectiveness of the overall method; adopts a deep neural network combined with a residual shrinkage unit to predict injury claims, and improves the original soft threshold filtering strategy of the residual shrinkage unit by constructing a combination layer and shrinking the filtering threshold function, which alleviates the problems of excessive filtering of information and loss of important information, improves the overall accuracy of the prediction, and improves the overall usability of intelligent prediction of injury claims.

Description

Technical Field

[0001] The present invention belongs to the technical field of claims prediction, and specifically refers to an intelligent prediction method and system for injury claims based on big data. Background Art

[0002] Intelligent prediction of injury claims aims to apply big data and artificial intelligence technology to automatically analyze injury claims related data and predict the claim amount, thereby significantly improving the speed and accuracy of claim processing, improving customer satisfaction, and helping to prevent unreasonable claims requests and enhance the risk management capabilities of insurance companies. However, in the existing intelligent prediction process of injury claims, there is a technical problem that the injury claim amount is inseparable from the injury level, and there is a lack of a method for predicting injury claims in combination with the injury level; there is an extremely complex prediction task for injury claims, involving a variety of data sources with high noise and inconsistent dimensions, and the traditional residual network with denoising capabilities often has excessive denoising in the prediction process, resulting in feature filtering that reduces the availability of information and loses important and more useful data information. Summary of the invention

[0003] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for intelligent prediction of injury claims based on big data. In view of the technical problem that the injury claim amount is inseparable from the injury level in the existing intelligent prediction process of injury claims, and there is a lack of a method for predicting injury claims in combination with the injury level, this solution creatively adopts a convolutional bidirectional long-short-term network combined with an attention mechanism to predict the injury level, better process complex data, enhance the robustness and accuracy of the model, effectively assist subsequent injury claims prediction, and improve the reliability and effectiveness of the overall method; in ... prediction in the existing intelligent prediction process of injury claims The prediction task is extremely complex, involving a variety of data sources with high noise and inconsistent dimensions. The traditional residual network with denoising capability often has excessive denoising during the prediction process, which leads to feature filtering that reduces the availability of information and loses important and more useful data information. This solution creatively uses a deep neural network combined with a residual shrinkage unit to predict injury claims, and improves the original soft threshold filtering strategy of the residual shrinkage unit by constructing a combination layer and shrinking the filtering threshold function, which alleviates the problems of excessive filtering of information and loss of important information, improves the overall accuracy of the prediction, and improves the overall availability of intelligent prediction of injury claims.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an intelligent prediction method for injury claims based on big data, the method comprising the following steps:

[0005] Step S1: data preparation;

[0006] Step S2: data preprocessing;

[0007] Step S3: Injury level prediction;

[0008] Step S4: injury claim prediction;

[0009] Step S5: Generate intelligent prediction report.

[0010] Furthermore, in step S1, the data preparation is used to prepare the original data required for intelligent prediction of injury claims, specifically collecting historical claims data, customer data, medical data and accident environment data to obtain the original injury data set;

[0011] The historical claims data includes claim amount, claim type, claim status and accident details;

[0012] The customer data, including the customer’s age, gender, occupation and health status;

[0013] The medical data, including medical diagnosis, treatment plan and treatment costs;

[0014] The accident environment data include the time, location and weather conditions of the accident.

[0015] Furthermore, in step S2, the data preprocessing is used to optimize the original data set, specifically, outliers and duplicate values ​​in the original injury data set are deleted, missing values ​​in the original injury data set are filled by the average value, and data normalization is performed to obtain an injury enhanced data set.

[0016] Further, in step S3, the injury level prediction is used to predict the injury level caused by the accident, specifically using a convolutional bidirectional long-term and short-term network combined with an attention mechanism to perform injury level prediction and obtain injury level information;

[0017] The convolutional bidirectional long-short-term network combined with the attention mechanism specifically includes an input layer, a one-dimensional convolutional network layer, a bidirectional long-short-term network layer, an attention mechanism, a fully connected layer and an output layer;

[0018] The one-dimensional convolutional network layer is used to extract local time series features;

[0019] The bidirectional long-term and short-term network layer is used to capture bidirectional dependencies;

[0020] The attention mechanism is used to enhance the model's focus on important features;

[0021] The injury level prediction includes the following steps:

[0022] Step S31: Constructing a convolutional bidirectional long-term short-term network combined with an attention mechanism, including the following steps:

[0023] Step S311: constructing an input layer, specifically receiving the injury enhancement data set as input data of the model through the input layer;

[0024] Step S312: construct a one-dimensional convolutional network layer, specifically, setting three convolutional combination layers, a flattening layer and a fully connected sublayer in the one-dimensional convolutional network layer, each convolutional combination layer includes a one-dimensional convolutional layer, a batch normalization sublayer and a random dropout sublayer, and calculating a one-dimensional convolutional feature vector through the one-dimensional convolutional network layer;

[0025] The calculation formula of the one-dimensional convolutional layer is:

[0026] ;

[0027] In the formula, cy i is the value of the output feature map at position i, where the output feature map specifically refers to the output of the one-dimensional convolution layer, i is the starting position index of the current sliding window, Relu(·) is the Relu activation function, j is the internal index of the convolution kernel, k is the convolution kernel size, specifically refers to the window size of the convolution kernel sliding on the input data, and w j is the weight of the convolution kernel at position j, x i+j is the value of the input data at position i+j, b c is the one-dimensional convolution bias term;

[0028] Step S313: constructing a bidirectional long-short term network layer, specifically setting a forward long-short term network layer, a backward long-short term network layer and a flattening layer in the bidirectional long-short term network layer, and calculating a bidirectional feature vector through the bidirectional long-short term network layer, including the following steps:

[0029] Step S3131: Construct the forward long-term and short-term network layer, and the calculation formula is:

[0030] ;

[0031] In the formula, is the hidden state of the forward long-term short-term network layer at the tth time step, t is the time step index, sig(·) is the S-type activation function, W1 is the first weight of the forward long-term short-term network layer, X t is the input feature vector at the tth time step, the input feature vector specifically refers to the one-dimensional convolution feature vector, W2 is the second weight of the forward long-term short-term network layer, is the hidden state of the forward long short-term network layer at the t-1th time step, tanh(·) is the hyperbolic tangent activation function, C t is the cell state of the forward long-term short-term network layer at the tth time step;

[0032] Step S3132: Construct the backward long-term and short-term network layer, and the calculation formula is:

[0033] ;

[0034] In the formula, is the hidden state of the backward long-term and short-term network layer at the tth time step, W3 is the first weight of the backward long-term and short-term network layer, W4 is the second weight of the backward long-term and short-term network layer, is the hidden state of the backward long short-term network layer at the t+1th time step, is the cell state of the backward long short-term network layer at the tth time step;

[0035] Step S3133: Hidden state splicing, the calculation formula is:

[0036] ;

[0037] In the formula, is the concatenated feature vector, W5 is the forward hidden state weight, and W6 is the backward hidden state weight;

[0038] Step S3134: construct a flattening layer to convert the concatenated features into a one-dimensional vector. The calculation formula is:

[0039] ;

[0040] In the formula, H bi is a bidirectional feature vector, specifically the output of the flatten layer, where flatten(·) is the flattening operation;

[0041] Step S3135: constructing the bidirectional long short-term network layer by constructing a forward long short-term network layer, constructing a backward long short-term network layer, concatenating hidden states, and constructing a flattening layer;

[0042] Step S314: construct an attention mechanism, specifically, apply the attention mechanism to calculate the attention feature vector of the bidirectional long-term and short-term network layer, and the calculation formula is:

[0043] ;

[0044] In the formula, A bi is the attention feature vector, T is the total number of time steps, is the attention weight at the tth time step, is the bidirectional feature vector at the tth time step, specifically the output feature vector of the bidirectional long-term and short-term network layer at the tth time step;

[0045] Step S315: construct a fully connected layer to generate prediction probability, and the calculation formula is:

[0046] ;

[0047] Where FCN is the predicted probability of the output of the fully connected layer, Softmax(·) is the Softmax function, and W fcn is the weight of the fully connected layer, b fcn is the fully connected layer bias term;

[0048] Step S316: constructing an output layer for outputting the prediction results of the model;

[0049] Step S32: constructing an injury level prediction model, specifically using a cross entropy loss function to perform model training on a convolutional bidirectional long-short term network combined with an attention mechanism to obtain an injury level prediction model;

[0050] Step S33: using the injury level prediction model to predict the injury level, and obtain injury level information, wherein the injury level information includes light injury, moderate injury, severe injury and critical injury.

[0051] Further, in step S4, the injury claim prediction is used to make a detailed prediction of the claim amount in combination with the injury level, specifically using a deep neural network combined with a residual shrinkage unit to perform injury claim prediction and obtain claim prediction information;

[0052] The deep neural network combined with the residual shrinkage unit specifically includes an input convolution layer, a residual shrinkage unit, a combination layer and a prediction output layer;

[0053] The input convolution layer is used to extract features and reduce training network parameters;

[0054] The residual shrinkage unit is used to optimize the problem of effective feature loss in the original residual network;

[0055] The combined layer is used to reduce the loss of discriminative information, and specifically includes a batch normalization layer, a Relu activation function, and an average pooling layer;

[0056] The injury claim prediction includes the following steps:

[0057] Step S41: Designing a deep neural network combined with a residual shrinkage unit, comprising the following steps:

[0058] Step S411: construct an input convolution layer, specifically accepting the injury level information and the original injury data set as the original input data samples of the model, and constructing a convolution layer to extract historical claims features, customer features, medical features, accident environment features and injury prediction features to obtain an input mapping feature data set;

[0059] The calculation formula for constructing the convolution layer and obtaining the input mapping feature data set is:

[0060] ;

[0061] In the formula, y in is the input mapping feature dataset, M is the total number of features, m is the feature index, x m is the mth feature of the input convolutional layer, k nm is the convolution weight corresponding to the mth feature data of the nth class, b n is the bias item corresponding to the nth type of feature, n is the feature category index, and the specific categories include historical claims features, customer features, medical features, accident environment features, and injury prediction features;

[0062] Step S412: construct a residual contraction unit, specifically, construct a residual contraction unit including a maximum pooling layer, a Relu activation layer, a batch normalization layer, a random dropout layer and two fully connected output layers, and construct a data filtering soft threshold to filter the feature data of the input mapping feature data set, so as to filter the more important data features for predicting injury claims, and obtain filtered claims feature data;

[0063] The data filtering soft threshold is specifically used to filter data through a soft threshold denoising function, and the calculation formula is:

[0064] ;

[0065] In the formula, y fil It is to filter the claim characteristic data. It is a positive filtering threshold parameter, which is determined by the average value of each feature data;

[0066] Step S413: constructing a residual contraction group, specifically, constructing seven residual contraction units in sequence after inputting the convolution layer to obtain a residual contraction group, and performing feature optimization filtering;

[0067] Step S414: Data filtering threshold optimization, specifically constructing a shrinkage filtering threshold function to replace the soft threshold denoising function in the original residual shrinkage unit, optimizing the filtering threshold, and obtaining optimized filtered claim feature data. The calculation formula is:

[0068] ;

[0069] In the formula, It is to optimize and filter the claim characteristic data. is the mth feature in the input mapping feature dataset, a m is the slope parameter of the shrinkage filter threshold function, and its specific value is {1, a m}, when the mth feature in the input mapping feature dataset and filter threshold parameters satisfy When the shrink filter threshold function slope parameter a m The value of , where e is a natural base. When the mth feature in the input mapping feature dataset and filter threshold parameters Dissatisfied When the shrink filter threshold function slope parameter a m The value of is 1;

[0070] Step S415: constructing a combination layer, specifically constructing a combination layer including a batch normalization layer, a Relu activation layer, and an average pooling layer, optimizing the feature normalization of the filtered claim feature data and reducing the loss of discriminant information, and obtaining claim prediction feature data;

[0071] Step S416: constructing a prediction output layer, specifically constructing a fully connected layer, and outputting a predicted value of the injury claim amount based on the claim prediction feature data through a classifier;

[0072] Step S42: constructing an injury claim prediction model, specifically using a cross entropy loss function to train a deep neural network combined with a residual shrinkage unit to obtain an injury claim prediction model;

[0073] Step S43: Perform injury claim prediction using the injury claim prediction model to obtain injury claim prediction information.

[0074] Furthermore, in step S5, the intelligent prediction report is generated, specifically, the intelligent prediction report is generated by integrating the injury claim prediction information and the injury level information.

[0075] The present invention provides an intelligent prediction system for injury claims based on big data, comprising: a data preparation module, a data preprocessing module, an injury level prediction module, an injury claims prediction module and an intelligent prediction report generation module;

[0076] The data preparation module is used for data preparation, and obtains an original injury data set through data preparation, and sends the original injury data set to the data preprocessing module and the injury claim prediction module;

[0077] The data preprocessing module is used for data preprocessing, obtaining an injury enhancement data set through data preprocessing, and sending the injury enhancement data set to the injury level prediction module;

[0078] The injury level prediction module is used for injury level prediction, obtains injury level information through injury level prediction, and sends the injury level information to the injury claim prediction module and the intelligent prediction report generation module;

[0079] The injury claim prediction module is used for injury claim prediction, obtains claim prediction information through injury claim prediction, and sends the claim prediction information to the intelligent prediction report generation module;

[0080] The intelligent prediction report generation module is used for intelligent prediction report generation, which integrates the claim prediction information and the injury level information to generate an intelligent prediction report.

[0081] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0082] (1) In view of the technical problems in the existing intelligent prediction process of injury claims, the amount of injury claims is inseparable from the injury level, and there is a lack of a method to predict injury claims based on the injury level. This solution creatively uses a convolutional bidirectional long-term and short-term network combined with an attention mechanism to predict the injury level, better handle complex data, enhance the robustness and accuracy of the model, effectively assist in subsequent injury claims prediction, and improve the reliability and effectiveness of the overall method.

[0083] (2) In the existing intelligent prediction process of injury claims, the prediction task of injury claims prediction is extremely complex, involving multiple data sources with high noise and inconsistent dimensions. The traditional residual network with denoising capability often has excessive denoising in the prediction process, which leads to feature filtering that reduces the availability of information and loses important and more useful data information. This solution creatively adopts a deep neural network combined with a residual shrinkage unit to predict injury claims, and improves the original soft threshold filtering strategy of the residual shrinkage unit by constructing a combination layer and a shrinkage filtering threshold function, which alleviates the problems of excessive filtering of information and loss of important information, improves the overall accuracy of the prediction, and improves the overall availability of intelligent prediction of injury claims. BRIEF DESCRIPTION OF THE DRAWINGS

[0084] Figure 1 A flowchart of an intelligent prediction method for injury claims based on big data provided by the present invention;

[0085] Figure 2 A schematic diagram of an intelligent injury claim prediction system based on big data provided by the present invention;

[0086] Figure 3 is a schematic flow chart of step S3;

[0087] Figure 4 It is a schematic diagram of the process of step S4.

[0088] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0089] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0090] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0091] Example 1, see Figure 1 The present invention provides an intelligent prediction method for injury claims based on big data, which includes the following steps:

[0092] Step S1: data preparation;

[0093] Step S2: data preprocessing;

[0094] Step S3: Injury level prediction;

[0095] Step S4: injury claim prediction;

[0096] Step S5: Generate intelligent prediction report.

[0097] Example 2, see Figure 1 , this embodiment is based on the above embodiment. In step S1, the data preparation is used to prepare the original data required for intelligent prediction of injury claims, specifically collecting historical claims data, customer data, medical data and accident environment data to obtain the original injury data set;

[0098] The historical claims data includes claim amount, claim type, claim status and accident details;

[0099] The customer data, including the customer’s age, gender, occupation and health status;

[0100] The medical data, including medical diagnosis, treatment plan and treatment costs;

[0101] The accident environment data include the time, location and weather conditions of the accident.

[0102] Example 3, see Figure 1This embodiment is based on the above embodiment. In step S2, the data preprocessing is used to optimize the original data set, specifically, outliers and duplicate values ​​in the original injury data set are deleted, missing values ​​in the original injury data set are filled with average values, and data normalization is performed to obtain an injury enhanced data set.

[0103] Example 4, see Figure 1 and Figure 3 This embodiment is based on the above embodiment. In step S3, the injury level prediction is used to predict the injury level caused by the accident. Specifically, a convolutional bidirectional long-term and short-term network combined with an attention mechanism is used to predict the injury level and obtain injury level information;

[0104] The convolutional bidirectional long-short-term network combined with the attention mechanism specifically includes an input layer, a one-dimensional convolutional network layer, a bidirectional long-short-term network layer, an attention mechanism, a fully connected layer and an output layer;

[0105] The one-dimensional convolutional network layer is used to extract local time series features;

[0106] The bidirectional long-term and short-term network layer is used to capture bidirectional dependencies;

[0107] The attention mechanism is used to enhance the model's focus on important features;

[0108] The injury level prediction includes the following steps:

[0109] Step S31: Constructing a convolutional bidirectional long-term short-term network combined with an attention mechanism, including the following steps:

[0110] Step S311: constructing an input layer, specifically receiving the injury enhancement data set as input data of the model through the input layer;

[0111] Step S312: construct a one-dimensional convolutional network layer, specifically, setting three convolutional combination layers, a flattening layer and a fully connected sublayer in the one-dimensional convolutional network layer, each convolutional combination layer includes a one-dimensional convolutional layer, a batch normalization sublayer and a random dropout sublayer, and calculating a one-dimensional convolutional feature vector through the one-dimensional convolutional network layer;

[0112] The calculation formula of the one-dimensional convolutional layer is:

[0113] ;

[0114] In the formula, cy iis the value of the output feature map at position i, where the output feature map specifically refers to the output of the one-dimensional convolution layer, i is the starting position index of the current sliding window, Relu(·) is the Relu activation function, j is the internal index of the convolution kernel, k is the convolution kernel size, specifically refers to the window size of the convolution kernel sliding on the input data, and w j is the weight of the convolution kernel at position j, x i+j is the value of the input data at position i+j, b c is the one-dimensional convolution bias term;

[0115] The one-dimensional convolution layer includes 64 convolution kernels of size 3;

[0116] Step S313: constructing a bidirectional long-short term network layer, specifically setting a forward long-short term network layer, a backward long-short term network layer and a flattening layer in the bidirectional long-short term network layer, and calculating a bidirectional feature vector through the bidirectional long-short term network layer, including the following steps:

[0117] Step S3131: Construct the forward long-term and short-term network layer, and the calculation formula is:

[0118] ;

[0119] In the formula, is the hidden state of the forward long-term short-term network layer at the tth time step, t is the time step index, sig(·) is the S-type activation function, W1 is the first weight of the forward long-term short-term network layer, X t is the input feature vector at the tth time step, the input feature vector specifically refers to the one-dimensional convolution feature vector, W2 is the second weight of the forward long-term short-term network layer, is the hidden state of the forward long short-term network layer at the t-1th time step, tanh(·) is the hyperbolic tangent activation function, C t is the cell state of the forward long-term short-term network layer at the tth time step;

[0120] Step S3132: Construct the backward long-term and short-term network layer, and the calculation formula is:

[0121] ;

[0122] In the formula, is the hidden state of the backward long-term and short-term network layer at the tth time step, W3 is the first weight of the backward long-term and short-term network layer, W4 is the second weight of the backward long-term and short-term network layer, is the hidden state of the backward long short-term network layer at the t+1th time step, is the cell state of the backward long short-term network layer at the tth time step;

[0123] Step S3133: Hidden state splicing, the calculation formula is:

[0124] ;

[0125] In the formula, is the concatenated feature vector, W5 is the forward hidden state weight, and W6 is the backward hidden state weight;

[0126] Step S3134: construct a flattening layer to convert the concatenated features into a one-dimensional vector. The calculation formula is:

[0127] ;

[0128] In the formula, H bi is a bidirectional feature vector, specifically the output of the flatten layer, where flatten(·) is the flattening operation;

[0129] Step S3135: constructing the bidirectional long short-term network layer by constructing a forward long short-term network layer, constructing a backward long short-term network layer, concatenating hidden states, and constructing a flattening layer;

[0130] Step S314: construct an attention mechanism, specifically, apply the attention mechanism to calculate the attention feature vector of the bidirectional long-term and short-term network layer, and the calculation formula is:

[0131] ;

[0132] In the formula, A bi is the attention feature vector, T is the total number of time steps, is the attention weight at the tth time step, is the bidirectional feature vector at the tth time step, specifically the output feature vector of the bidirectional long-term and short-term network layer at the tth time step;

[0133] Step S315: construct a fully connected layer to generate prediction probability, and the calculation formula is:

[0134] ;

[0135] Where FCN is the predicted probability of the output of the fully connected layer, Softmax(·) is the Softmax function, and W fcn is the weight of the fully connected layer, b fcn is the fully connected layer bias term;

[0136] Step S316: constructing an output layer for outputting the prediction results of the model;

[0137] Step S32: constructing an injury level prediction model, specifically using a cross entropy loss function to perform model training on a convolutional bidirectional long-short term network combined with an attention mechanism to obtain an injury level prediction model;

[0138] Step S33: using an injury level prediction model to predict the injury level, and obtaining injury level information, wherein the injury level information includes light injury, moderate injury, severe injury, and critical injury;

[0139] By performing the above operations, in order to address the technical problem that in the existing intelligent prediction process of injury claims, the injury claim amount is inseparable from the injury level, and there is a lack of a method to predict injury claims based on the injury level, this solution creatively adopts a convolutional bidirectional long-short term network combined with an attention mechanism to predict the injury level, better handle complex data, enhance the robustness and accuracy of the model, effectively assist in subsequent injury claims predictions, and improve the reliability and effectiveness of the overall method.

[0140] Example 5, see Figure 1 and Figure 4 This embodiment is based on the above embodiment. In step S4, the injury claim prediction is used to make a detailed prediction of the claim amount in combination with the injury level. Specifically, a deep neural network combined with a residual shrinkage unit is used to perform injury claim prediction to obtain claim prediction information;

[0141] The deep neural network combined with the residual shrinkage unit specifically includes an input convolution layer, a residual shrinkage unit, a combination layer and a prediction output layer;

[0142] The input convolution layer is used to extract features and reduce training network parameters;

[0143] The residual shrinkage unit is used to optimize the problem of effective feature loss in the original residual network;

[0144] The combined layer is used to reduce the loss of discriminative information, and specifically includes a batch normalization layer, a Relu activation function, and an average pooling layer;

[0145] The injury claim prediction includes the following steps:

[0146] Step S41: Designing a deep neural network combined with a residual shrinkage unit, comprising the following steps:

[0147] Step S411: construct an input convolution layer, specifically accepting the injury level information and the original injury data set as the original input data samples of the model, and constructing a convolution layer to extract historical claims features, customer features, medical features, accident environment features and injury prediction features to obtain an input mapping feature data set;

[0148] The calculation formula for constructing the convolution layer and obtaining the input mapping feature data set is:

[0149] ;

[0150] In the formula, yin is the input mapping feature dataset, M is the total number of features, m is the feature index, x m is the mth feature of the input convolutional layer, k nm is the convolution weight corresponding to the mth feature data of the nth class, b n is the bias item corresponding to the nth type of feature, n is the feature category index, and the specific categories include historical claims features, customer features, medical features, accident environment features, and injury prediction features;

[0151] Step S412: construct a residual contraction unit, specifically, construct a residual contraction unit including a maximum pooling layer, a Relu activation layer, a batch normalization layer, a random dropout layer and two fully connected output layers, and construct a data filtering soft threshold to filter the feature data of the input mapping feature data set, so as to filter the more important data features for predicting injury claims, and obtain filtered claims feature data;

[0152] The data filtering soft threshold is specifically used to filter data through a soft threshold denoising function, and the calculation formula is:

[0153] ;

[0154] In the formula, y fil It is to filter the claim characteristic data. It is a positive filtering threshold parameter, which is determined by the average value of each feature data;

[0155] Step S413: constructing a residual contraction group, specifically, constructing seven residual contraction units in sequence after inputting the convolution layer to obtain a residual contraction group, and performing feature optimization filtering;

[0156] The plurality of residual shrinkage units are specifically seven residual shrinkage units;

[0157] Step S414: Data filtering threshold optimization, specifically constructing a shrinkage filtering threshold function to replace the soft threshold denoising function in the original residual shrinkage unit, optimizing the filtering threshold, and obtaining optimized filtered claim feature data. The calculation formula is:

[0158] ;

[0159] In the formula, It is to optimize and filter the claim characteristic data. is the mth feature in the input mapping feature dataset, a m is the slope parameter of the shrinkage filter threshold function, and its specific value is {1, a m}, when the mth feature in the input mapping feature dataset and filter threshold parameters satisfy When the shrink filter threshold function slope parameter a m The value of , where e is a natural base. When the mth feature in the input mapping feature dataset and filter threshold parameters Dissatisfied When the shrink filter threshold function slope parameter a m The value of is 1;

[0160] Step S415: constructing a combination layer, specifically constructing a combination layer including a batch normalization layer, a Relu activation layer, and an average pooling layer, optimizing the feature normalization of the filtered claim feature data and reducing the loss of discriminant information, and obtaining claim prediction feature data;

[0161] Step S416: constructing a prediction output layer, specifically constructing a fully connected layer, and outputting a predicted value of the injury claim amount based on the claim prediction feature data through a classifier;

[0162] Step S42: constructing an injury claim prediction model, specifically using a cross entropy loss function to train a deep neural network combined with a residual shrinkage unit to obtain an injury claim prediction model;

[0163] Step S43: performing injury claim prediction using an injury claim prediction model to obtain injury claim prediction information;

[0164] By performing the above operations, in the existing intelligent prediction process of injury claims, the prediction task of injury claims prediction is extremely complex, involving multiple data sources with high noise and inconsistent dimensions, and the traditional residual network with denoising capability often has excessive denoising in the prediction process, resulting in feature filtering that reduces the availability of information and loses important and more useful data information. Technical problems, this solution creatively adopts a deep neural network combined with a residual shrinkage unit to predict injury claims, and improves the original soft threshold filtering strategy of the residual shrinkage unit by constructing a combination layer and shrinking the filtering threshold function, which alleviates the problems of excessive filtering of information and loss of important information, improves the overall accuracy of the prediction, and improves the overall availability of intelligent prediction of injury claims.

[0165] Example 6, see Figure 1 This embodiment is based on the above embodiment. In step S5, the intelligent prediction report is generated, specifically, the intelligent prediction report is generated by integrating the injury claim prediction information and the injury level information.

[0166] Embodiment 7, see Figure 2, this embodiment is based on the above embodiment, and the present invention provides an intelligent prediction system for injury claims based on big data, including: a data preparation module, a data preprocessing module, an injury level prediction module, an injury claims prediction module and an intelligent prediction report generation module;

[0167] The data preparation module is used for data preparation, and obtains an original injury data set through data preparation, and sends the original injury data set to the data preprocessing module and the injury claim prediction module;

[0168] The data preprocessing module is used for data preprocessing, obtaining an injury enhancement data set through data preprocessing, and sending the injury enhancement data set to the injury level prediction module;

[0169] The injury level prediction module is used for injury level prediction, obtains injury level information through injury level prediction, and sends the injury level information to the injury claim prediction module and the intelligent prediction report generation module;

[0170] The injury claim prediction module is used for injury claim prediction, obtains claim prediction information through injury claim prediction, and sends the claim prediction information to the intelligent prediction report generation module;

[0171] The intelligent prediction report generation module is used for intelligent prediction report generation, which integrates the claim prediction information and the injury level information to generate an intelligent prediction report.

[0172] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device 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 device.

[0173] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0174] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. An intelligent prediction method for injury claims based on big data, characterized by: The method comprises the following steps: Step S1: Data preparation, which is used to prepare the original data required for intelligent prediction of injury claims, specifically collecting historical claims data, customer data, medical data and accident environment data to obtain the original injury data set; The historical claims data includes claim amount, claim type, claim status and accident details; The customer data, including the customer’s age, gender, occupation and health status; The medical data, including medical diagnosis, treatment plan and treatment costs; The accident environment data, including the time, location and weather conditions of the accident; Step S2: data preprocessing; Step S3: Injury level prediction, used to predict the injury level caused by the accident, specifically using a convolutional bidirectional long-term and short-term network combined with an attention mechanism to predict the injury level and obtain injury level information, wherein the injury level information includes light injury, moderate injury, severe injury and critical injury; The convolutional bidirectional long-short-term network combined with the attention mechanism specifically includes an input layer, a one-dimensional convolutional network layer, a bidirectional long-short-term network layer, an attention mechanism, a fully connected layer and an output layer; The one-dimensional convolutional network layer is used to extract local time series features; The bidirectional long-term and short-term network layer is used to capture bidirectional dependencies; The attention mechanism is used to enhance the model's focus on important features; Step S4: Injury claim prediction, which is used to make a detailed prediction of the claim amount based on the injury level, and adopt a deep neural network combined with a residual shrinkage unit to make injury claim prediction and obtain claim prediction information; The deep neural network combined with the residual shrinkage unit specifically includes an input convolution layer, a residual shrinkage unit, a combination layer and a prediction output layer; The input convolution layer is used to extract features and reduce training network parameters; The residual shrinkage unit is used to optimize the problem of effective feature loss in the original residual network; The combined layer is used to reduce the loss of discriminative information, and specifically includes a batch normalization layer, a Relu activation function, and an average pooling layer; The injury claim prediction includes the following steps: Step S41: designing a deep neural network combined with a residual shrinkage unit, which is used to improve the original soft threshold filtering strategy of the residual shrinkage unit by constructing a combination layer and shrinking the filtering threshold function, thereby alleviating the problems of excessive filtering of information and loss of important information, including the following steps: Step S411: construct an input convolution layer, specifically accepting the injury level information and the original injury data set as the original input data samples of the model, and constructing a convolution layer to extract historical claims features, customer features, medical features, accident environment features and injury prediction features to obtain an input mapping feature data set; The calculation formula for constructing the convolution layer and obtaining the input mapping feature data set is: ； In the formula, y in is the input mapping feature dataset, M is the total number of features, m is the feature index, x m is the mth feature of the input convolutional layer, k nm is the convolution weight corresponding to the mth feature data of the nth class, b n is the bias item corresponding to the nth type of feature, n is the feature category index, and the specific categories include historical claims features, customer features, medical features, accident environment features, and injury prediction features; Step S412: construct a residual contraction unit, specifically, construct a residual contraction unit including a maximum pooling layer, a Relu activation layer, a batch normalization layer, a random dropout layer and two fully connected output layers, and construct a data filtering soft threshold to filter the feature data of the input mapping feature data set, so as to filter the more important data features for predicting injury claims, and obtain filtered claims feature data; The data filtering soft threshold is specifically used to filter data through a soft threshold denoising function, and the calculation formula is: ； In the formula, y fil It is to filter the claim characteristic data. It is a positive filtering threshold parameter, which is determined by the average value of each feature data; Step S413: constructing a residual contraction group, specifically, constructing seven residual contraction units in sequence after inputting the convolution layer to obtain a residual contraction group, and performing feature optimization filtering; Step S414: Data filtering threshold optimization, specifically constructing a shrinkage filtering threshold function to replace the soft threshold denoising function in the original residual shrinkage unit, optimizing the filtering threshold, and obtaining optimized filtered claim feature data. The calculation formula is: ； In the formula, It is to optimize and filter the claim characteristic data. is the mth feature in the input mapping feature dataset, a m is the slope parameter of the shrinkage filter threshold function, and its specific value is {1, a m }, when the mth feature in the input mapping feature dataset and filter threshold parameters satisfy When the shrink filter threshold function slope parameter a m The value of , where e is a natural base. When the mth feature in the input mapping feature dataset and filter threshold parameters Dissatisfied When the shrink filter threshold function slope parameter a m The value of is 1; Step S415: constructing a combination layer, specifically constructing a combination layer including a batch normalization layer, a Relu activation layer, and an average pooling layer, optimizing the feature normalization of the filtered claim feature data and reducing the loss of discriminant information, and obtaining claim prediction feature data; Step S416: constructing a prediction output layer, specifically constructing a fully connected layer, and outputting a predicted value of the injury claim amount based on the claim prediction feature data through a classifier; Step S42: constructing an injury claim prediction model, specifically using a cross entropy loss function to train a deep neural network combined with a residual shrinkage unit to obtain an injury claim prediction model; Step S43: performing injury claim prediction using an injury claim prediction model to obtain injury claim prediction information; Step S5: Generate intelligent prediction report.

2. According to the big data-based intelligent prediction method for injury claims according to claim 1, it is characterized by: In step S3, the injury level prediction includes the following steps: Step S31: Constructing a convolutional bidirectional long-term short-term network combined with an attention mechanism, including the following steps: Step S311: constructing an input layer, specifically receiving the injury enhancement data set as input data of the model through the input layer; Step S312: construct a one-dimensional convolutional network layer, specifically, setting three convolutional combination layers, a flattening layer and a fully connected sublayer in the one-dimensional convolutional network layer, each convolutional combination layer includes a one-dimensional convolutional layer, a batch normalization sublayer and a random dropout sublayer, and calculating a one-dimensional convolutional feature vector through the one-dimensional convolutional network layer; The calculation formula of the one-dimensional convolutional layer is: ； In the formula, cy i is the value of the output feature map at position i, where the output feature map specifically refers to the output of the one-dimensional convolution layer, i is the starting position index of the current sliding window, Relu(·) is the Relu activation function, j is the internal index of the convolution kernel, k is the convolution kernel size, specifically refers to the window size of the convolution kernel sliding on the input data, and w j is the weight of the convolution kernel at position j, x i+j is the value of the input data at position i+j, b c is the one-dimensional convolution bias term; Step S313: constructing a bidirectional long-short term network layer, specifically setting a forward long-short term network layer, a backward long-short term network layer and a flattening layer in the bidirectional long-short term network layer, and calculating a bidirectional feature vector through the bidirectional long-short term network layer, including the following steps: Step S3131: Construct the forward long-term and short-term network layer, and the calculation formula is: ； In the formula, is the hidden state of the forward long-term short-term network layer at the tth time step, t is the time step index, sig(·) is the S-type activation function, W1 is the first weight of the forward long-term short-term network layer, X t is the input feature vector at the tth time step, the input feature vector specifically refers to the one-dimensional convolution feature vector, W2 is the second weight of the forward long-term short-term network layer, is the hidden state of the forward long short-term network layer at the t-1th time step, tanh(·) is the hyperbolic tangent activation function, C t is the cell state of the forward long-term short-term network layer at the tth time step; Step S3132: Construct the backward long-term and short-term network layer, and the calculation formula is: ； In the formula, is the hidden state of the backward long-term and short-term network layer at the tth time step, W3 is the first weight of the backward long-term and short-term network layer, W4 is the second weight of the backward long-term and short-term network layer, is the hidden state of the backward long short-term network layer at the t+1th time step, is the cell state of the backward long short-term network layer at the tth time step; Step S3133: Hidden state splicing, the calculation formula is: ； In the formula, is the concatenated feature vector, W5 is the forward hidden state weight, and W6 is the backward hidden state weight; Step S3134: construct a flattening layer to convert the concatenated features into a one-dimensional vector. The calculation formula is: ； In the formula, H bi is a bidirectional feature vector, specifically the output of the flatten layer, where flatten(·) is the flattening operation; Step S3135: constructing the bidirectional long short-term network layer by constructing a forward long short-term network layer, constructing a backward long short-term network layer, concatenating hidden states, and constructing a flattening layer; Step S314: construct an attention mechanism, specifically, apply the attention mechanism to calculate the attention feature vector of the bidirectional long-term and short-term network layer, and the calculation formula is: ； In the formula, A bi is the attention feature vector, T is the total number of time steps, is the attention weight at the tth time step, is the bidirectional feature vector at the tth time step, specifically the output feature vector of the bidirectional long-term and short-term network layer at the tth time step; Step S315: construct a fully connected layer to generate prediction probability, and the calculation formula is: ； Where FCN is the predicted probability of the output of the fully connected layer, Softmax(·) is the Softmax function, and W fcn is the weight of the fully connected layer, b fcn is the fully connected layer bias term; Step S316: constructing an output layer for outputting the prediction results of the model; Step S32: constructing an injury level prediction model, specifically using a cross entropy loss function to perform model training on a convolutional bidirectional long-short term network combined with an attention mechanism to obtain an injury level prediction model; Step S33: using the injury level prediction model to predict the injury level and obtain injury level information.

3. According to the big data-based intelligent prediction method for injury claims according to claim 2, it is characterized by: In step S2, the data preprocessing is used to optimize the original data set, specifically, outliers and duplicate values ​​in the original injury data set are deleted, missing values ​​in the original injury data set are filled by the average value, and data normalization is performed to obtain an injury enhanced data set; In step S5, the intelligent prediction report is generated, specifically, the intelligent prediction report is generated by integrating the injury claim prediction information and the injury level information.

4. An intelligent prediction system for injury claims based on big data, used to implement an intelligent prediction method for injury claims based on big data as described in any one of claims 1 to 3, characterized in that: It includes data preparation module, data preprocessing module, injury level prediction module, injury claim prediction module and intelligent prediction report generation module; The data preparation module obtains an original injury data set through data preparation, and sends the original injury data set to the data preprocessing module and the injury claim prediction module; The data preprocessing module obtains an injury enhancement data set through data preprocessing, and sends the injury enhancement data set to the injury level prediction module; The injury level prediction module obtains injury level information through injury level prediction, and sends the injury level information to the injury claim prediction module and the intelligent prediction report generation module; The injury claim prediction module obtains claim prediction information through injury claim prediction, and sends the claim prediction information to the intelligent prediction report generation module; The intelligent prediction report generation module generates an intelligent prediction report by integrating the claim prediction information and the injury level information.

Images