Risk control rule effect prediction method and system based on big data

Abstract

The invention discloses a risk control rule effect prediction method and system based on big data, and relates to the technical field of financial risk control, and the method comprises the steps: collecting a transaction detail data set, a user activity record data set, an insurance claim data set, a geographic position data set and a time stamp information data set from a multivariate data source, preprocessing the collected information to form a data set; constructing an anomaly detection model by using a generative adversarial network GANs, and identifying an abnormal behavior in the transaction mode; constructing a risk control prediction model based on a space-time diagram convolutional neural network STGCN, and inputting the abnormal behavior data in combination with risk control rule parameters into the risk control prediction model to obtain a prediction risk probability; comprehensively evaluating the transaction behavior based on the abnormal behavior in combination with insurance claim evaluation, fraud detection and insurance market fluctuation to obtain a comprehensive evaluation result; a corresponding risk control prevention strategy is made based on the prediction risk probability and the comprehensive evaluation result; and inputting the real-time data into the model to obtain a real-time risk control prevention strategy, and executing the strategy.

Description

Technical Field

[0001] The present invention relates to the technical field of financial risk control, and particularly to a method and system for predicting the effect of risk control rules based on big data. Background Art

[0002] In the context of the Internet of Vehicles and big data, the auto insurance industry is undergoing an innovation in risk control means. The traditional claims review is inefficient and error-prone. Intelligent risk control solutions have emerged, using deep learning technologies GANs and STGCN to integrate multi-source data on vehicle accidents and driving behaviors, identify fraud, and predict risks. Customers seek flexible risk control rule adjustments, which prompts us to develop a big data prediction model to evaluate the impact of rule changes on premiums, costs, and customer satisfaction. At the same time, the automated loss assessment system integrates historical and real-time data, automatically generates accurate reports, accelerates claims settlement, reduces errors, and optimizes risk management.

[0003] However, the main deficiencies faced by the existing technologies include the timeliness of data processing, the interpretability of models, and the flexibility of rule adjustments. Traditional risk control models may not be able to analyze massive data streams in real time, resulting in decision-making lags. The black-box characteristics of deep learning models reduce their transparency in the claims settlement process, affecting customer trust. In addition, the current risk control rule adjustments often rely on manual experience and lack effective quantitative tools to evaluate the impact of rule changes. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for predicting the effect of risk control rules based on big data to solve the problems of the timeliness of data processing, the interpretability of models, and the flexibility of rule adjustments. Traditional risk control models may not be able to analyze massive data streams in real time, resulting in decision-making lags. The "black-box" characteristics of deep learning models reduce their transparency in the claims settlement process, affecting customer trust. In addition, the current risk control rule adjustments often rely on manual experience and lack effective quantitative tools to evaluate the impact of rule changes.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a method for predicting the effect of risk control rules based on big data, which includes collecting a transaction details data set, a user activity record data set, an insurance claims data set, a geographical location data set, and a time stamp information data set from multi-source data sources, preprocessing the collected information to form a data set;

[0008] Using the generative adversarial network GANs to construct an anomaly detection model to identify abnormal behaviors in transaction patterns;

[0009] Build a risk control prediction model based on the spatio-temporal graph convolutional neural network STGCN, input the abnormal behavior data combined with the risk control rule parameters into the risk control prediction model, and obtain the predicted risk probability;

[0010] Based on the abnormal behavior, combined with insurance claim assessment, fraud detection, and insurance market fluctuations, comprehensively evaluate the transaction behavior to obtain a comprehensive evaluation result;

[0011] Based on the predicted risk probability and the comprehensive evaluation result, formulate corresponding risk control prevention strategies;

[0012] Input the real-time data into the model to obtain the real-time risk control prevention strategy and execute the strategy.

[0013] As a preferred solution of the risk control rule effect prediction method based on big data described in the present invention, wherein: collecting the transaction details dataset, user activity record dataset, insurance claim dataset, geographical location dataset, and time stamp information dataset from multiple data sources, and preprocessing the collected information to form a dataset. The specific steps are as follows:

[0014] Set the original data sets of the transaction details dataset, user activity record dataset, insurance claim dataset, geographical location dataset, and time stamp information dataset as D, and the expression is:

[0015]

[0016] Among them, T, U, M, G, and Tm respectively represent the transaction details dataset, user activity record dataset, insurance claim dataset, geographical location dataset, and time stamp information dataset;

[0017] Each dataset contains its own data points, and the expression is:

[0018]

[0019] Among them, X i represents the i-th dataset, and each dataset contains multiple data points X ij , i represents the data type, and j represents the data point number;

[0020] For the transaction details dataset, convert the transaction details data, and the expression is:

[0021] f(T) = log(1 + T);

[0022] Among them, f(T) represents the conversion of the transaction details dataset T, log represents the natural logarithm function, and T a represents the transaction amount;

[0023] Preprocess the user activity record dataset, and the expression is:

[0024]

[0025] Among them, f(U) represents the value of the normalized user activity record dataset U, min(U) represents the minimum value in the dataset, and max(U) represents the maximum value in the dataset;

[0026] For the insurance claim dataset, obtain its periodic change pattern, and the expression is:

[0027] g(M) = sin(ω M ·t + φ M );

[0028] Among them, g(M) represents the influence of market dynamics changing over time, sin represents the sine function, w M represents the angular frequency of the insurance claim dataset, t represents the time point, represents the phase shift of the insurance claim dataset;

[0029] For the geographical location dataset, preprocess it, and the expression is:

[0030] h(G) = exp(-α·d 2 );

[0031] Among them, h(G) represents the influence of the trading location on the trading risk, exp represents the exponential function, α represents the attenuation coefficient, and d represents the Euclidean distance between two points;

[0032] For the time stamp information dataset, preprocess it, and the expression is:

[0033]

[0034] Among them, k(Tm) represents the influence of an event over time, e represents the base of the natural logarithm, β represents the steepness parameter, t represents the time point, and τ represents the time threshold;

[0035] Fuse the preprocessed data to form a dataset, and the expression is:

[0036]

[0037] Among them, D ′ represents the preprocessed dataset.

[0038] As a preferred solution of the risk control rule effect prediction method based on big data according to the present invention, wherein: the abnormal detection model is constructed by using the generative adversarial network GANs to identify abnormal behaviors in the trading pattern, and the specific steps are:

[0039] Set the generator network as G, the discriminator network as DL, X as the real transaction data, and Z as the random noise vector;

[0040] In the generator and discriminator networks of GANs, simultaneously take the risk control rule parameters as part of the input features, use one-hot encoding to convert these parameters into numerical form, and then input them into the anomaly detection model together with the transaction data;

[0041] Use the loss function to measure the difference between the real transaction data and the synthetic data. The expression is:

[0042]

[0043] Among them, L GAN (G, DL) represents the loss function of the generative adversarial network, P d represents the distribution of the real data, P n represents the distribution of the noise, represents the average of the variable x under the real data, represents the average of the variable z under the noise distribution, log(DL(x)) represents the output of the discriminator DL for the real data x converted by the natural logarithm function;

[0044] Set the anomaly scoring function to identify abnormal behaviors in the transaction pattern. The expression is:

[0045]

[0046] Among them, S a represents the anomaly scoring function, which is used to quantify the anomaly degree of the transaction data point x ′ , x ′ represents the preprocessed transaction record, α is a constant, μ represents the average value of the output of the discriminator DL for the real transaction data during training, and e represents the base of the natural logarithm;

[0047] Set the threshold IO;

[0048] The value range of the anomaly scoring function S a is from 0 to 1. When 0 ≤ S a < IO, it means that the transaction pattern is highly consistent with the normal behavior. When IO ≤ S a ≤ 1, it means that the transaction pattern significantly deviates from the normal behavior and is an abnormal behavior.

[0049] As a preferred solution of the risk control rule effect prediction method based on big data according to the present invention, wherein: construct a risk control prediction model based on the spatio-temporal graph convolutional neural network STGCN, input the abnormal behavior data into the risk control prediction model, and obtain the predicted risk probability. The specific steps are as follows:

[0050] Based on the preprocessed dataset D ′ Construct a spatio-temporal graph, and extract transaction details, user activities, market dynamics, geographical locations, and time stamp information from D ′ to form the features of the nodes, that is, the positions and attributes of the nodes in the spatio-temporal graph. Each node v represents a specific spatio-temporal event. All node sets are formed into a node set V. Based on the node set and the edge set, a spatio-temporal graph is generated. The expression is:

[0051] G st =(V, E);

[0052] where G st represents the spatio-temporal graph, s represents the geographical coordinates, t represents the time stamp, V represents the node set in the spatio-temporal graph, and E represents the edge set in the spatio-temporal graph;

[0053] Set the spatio-temporal adjacency matrix as A. The spatio-temporal adjacency matrix A is a two-dimensional matrix, where the rows and columns respectively correspond to each node in the graph. If there is a direct connection between i and j, then the element a ij in A will be a non-zero value, indicating their degree of association; if there is no direct connection, then the value of a ij is 0;

[0054] Use the Gaussian kernel function K to calculate the element aij in the adjacency matrix. The expression is:

[0055]

[0056] where a ij represents the adjacency matrix element between node i and node j, α is the time decay coefficient, d t ij represents the distance between node i and node j in the time dimension, β represents the space decay coefficient, and d s ij represents the distance between node i and node j in the space dimension;

[0057] In STGCN, use the spatio-temporal graph convolutional layer to propagate information. The expression of the new hidden state after being processed by the spatio-temporal graph convolutional layer is:

[0058]

[0059] where h v new represents the new hidden state of node v after being processed by the spatio-temporal graph convolutional layer, σ is the activation function, W s represents the spatial weight matrix, a vu represents the adjacency weight between node v and node u, and h u represents the current hidden state of node ut denotes the time-weighted matrix, h v represents the current hidden state of node v;

[0060] Set the original feature vector of each node v in the spatio-temporal graph as x v ,

[0061] Attach the anomaly score S a to the feature vector x v to form an enhanced feature vector, and the expression is:

[0062]

[0063] where x v avg represents the enhanced feature vector;

[0064] Input all the enhanced feature vectors into the risk control prediction model to complete the training, and update the hidden state of each node with the enhanced feature vector x v avg After processing through the multi-layer STGCN, obtain the final hidden state h of each node v f ;

[0065] Pass the hidden state to the fully connected layer to predict the risk probability, and the output layer expression is:

[0066]

[0067] where p r represents the predicted risk probability, σ represents the activation function, and F represents the fully connected layer;

[0068] Set the threshold Υ;

[0069] The value range of the predicted risk probability p r is from 0 to 1. When the value of p r is less than Υ, it indicates low risk. When the value of p r is greater than or equal to Υ, it indicates high risk.

[0070] As a preferred solution of the method for predicting the effect of risk control rules based on big data according to the present invention, wherein: the transaction behavior is comprehensively evaluated by combining abnormal behavior with insurance claim assessment, fraud detection, and insurance market fluctuations to obtain a comprehensive evaluation result. The specific steps are as follows:

[0071] Collect historical claim data from historical data;

[0072] Preprocess the claim historical data;

[0073] Use the deep learning model LSTM to construct a claim settlement habit prediction model P based on historical claim data ins , and use the historical claim data to input the claim settlement habit prediction model to complete the training;

[0074] Use the cross-validation method to verify the output score of the model and optimize the parameters to improve the generalization ability and prediction accuracy of the model;

[0075] Based on the output score of the claim settlement habit prediction model, calculate the claim habit score, and the expression is:

[0076] I s = f(P ins (Ru), We Ins );

[0077] Among them, I s is the claim habit score, f represents the claim function, Ru represents the risk control rule parameter, and We ins represents the weight vector of the claim habit score;

[0078] Based on the user's historical credit performance and payment behavior, calculate the insurance claim assessment score, and the expression is:

[0079]

[0080] Among them, C s represents the insurance claim assessment score, H represents the user's historical credit score, μ H represents the historical average credit score, D represents the average payment delay days, e represents the base of the natural logarithm, and w1 and w2 are the weight coefficients of the user's historical credit score and average payment delay days respectively;

[0081] Based on the anomaly detection model, calculate the fraud detection score, and the expression is:

[0082]

[0083] Among them, F d represents the fraud detection score, S min represents the minimum value of the anomaly score, and S max represents the maximum value of the anomaly score;

[0084] Based on the periodic change pattern in the insurance claim data set M, calculate the insurance market volatility score, and the expression is:

[0085]

[0086] Among them, σ represents the volatility index of the stock price, and σ min and σ max represent the minimum and maximum values of the volatility index respectively,

[0087] The claim settlement habit score, together with the insurance claim settlement assessment score, fraud detection score, and insurance market volatility score, is input into the calculation formula for the comprehensive assessment result, and the expression is:

[0088]

[0089] Among them, R c represents the comprehensive assessment result, and W c , W f , W m , W i respectively represent the weights of the insurance claim settlement assessment score, fraud detection score, insurance market volatility score, and claim settlement habit score. λ1, λ2, λ3, and λ4 represent adjustment factors used to control the degree of non-linear influence on the comprehensive assessment result R c when each score deviates from its benchmark value, and μ C , μ F , μ M , μ I are respectively the average values of the insurance claim settlement assessment score, fraud detection score, insurance market volatility score, and claim settlement habit score.

[0090] As an optimal solution of the risk control rule effect prediction method based on big data according to the present invention, wherein: based on the predicted risk probability and the comprehensive assessment result, corresponding risk control prevention strategies are formulated, and the specific steps are as follows:

[0091] Set the market sentiment index as M e ;

[0092] Combined with the predicted risk probability p r , the comprehensive assessment result R c and the market sentiment index M e to construct a decision function, and the expression is:

[0093]

[0094] Among them, D f represents the decision function score, γ represents the market sentiment adjustment factor, δ represents the non-linear influence index of the predicted risk probability p r , η represents the non-linear influence index of the comprehensive assessment result R c , and ξ represents the balance coefficient;

[0095] The value range of the decision function score D f is from 0 to 1;

[0096] When D f is close to 0, it indicates that the transaction risk is relatively low, and fewer risk control measures can be taken;

[0097] When Df When it is close to 1, it indicates a high trading risk and strict risk control measures need to be taken.

[0098] As an optimal solution of the method for predicting the effect of risk control rules based on big data according to the present invention, wherein: inputting the real-time data into the model to obtain a real-time risk control prevention strategy and execute the strategy, the specific steps are as follows:

[0099] Collect real-time data from multiple data sources;

[0100] Preprocess the real-time data respectively, and then input it into the anomaly detection model to obtain the real-time anomaly score;

[0101] Combine the real-time anomaly score with the comprehensive evaluation result, input it into the risk control prediction model for real-time risk assessment, and output the real-time predicted risk probability;

[0102] Calculate the real-time comprehensive evaluation result based on the real-time anomaly score;

[0103] Input the real-time predicted risk probability and the real-time comprehensive evaluation result into the decision function formula to obtain the real-time decision function score, and at the same time execute the corresponding risk control strategy.

[0104] In a second aspect, the present invention provides a system for predicting the effect of risk control rules based on big data, including,

[0105] An acquisition module, responsible for collecting transaction details, user activities, market dynamics, geographical locations and time stamp information from multiple data sources, preprocessing these data, and converting them into a format available for model analysis;

[0106] A detection and scoring module, responsible for using generative adversarial networks (GANs) to identify abnormal behaviors in trading patterns, assigning anomaly scores to each trading data point, and quantifying the degree of deviation from normal behavior;

[0107] A risk prediction module, predicting the risk probability of a single transaction based on a spatio-temporal graph convolutional neural network (STGCN), and conducting a comprehensive evaluation by combining factors such as anomaly scores, insurance claim assessments, fraud detections, and insurance market fluctuations;

[0108] A decision-making module, responsible for combining the predicted risk probability, the comprehensive evaluation result, and the market sentiment index to generate a decision function score, and formulating corresponding risk control strategies according to the decision function score;

[0109] An execution module, responsible for receiving and processing new data in real time, dynamically evaluating risks, and immediately executing the corresponding risk control strategies.

[0110] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, where the memory stores a computer program, and: when the computer program is executed by the processor, any step of the risk control rule effect prediction method based on big data as described in the first aspect of the present invention is implemented.

[0111] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and: when the computer program is executed by the processor, any step of the risk control rule effect prediction method based on big data as described in the first aspect of the present invention is implemented.

[0112] The beneficial effects of the present invention are as follows: through the input of real-time data, the instantaneity and accuracy of the risk control strategy are ensured; by optimizing the model structure and algorithm, the interpretability of model decision-making is enhanced, and customers' understanding and trust in the claims settlement process are improved; the risk control rule effect prediction model provides a scientific basis for rule adjustment, enabling insurance companies to make decisions based on quantitative analysis, thereby improving the efficiency and effect of rule adjustment. Description of the Drawings

[0113] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for description in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0114] Figure 1 It is a flowchart of the risk control rule effect prediction method based on big data in Embodiment 1.

[0115] Figure 2 It is a flowchart of the score interval of the decision function in Embodiment 1. Detailed Embodiments

[0116] To make the above objects, features, and advantages of the present invention more obvious and understandable, the detailed embodiments of the present invention will be described in detail below with reference to the drawings in the specification.

[0117] Many specific details are set forth in the following description in order to provide a thorough understanding of the present invention. However, the present invention may be implemented in other ways different from those described herein. Those skilled in the art can make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0118] Second, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation manner of the present invention. The phrase "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor is it an individual or alternative embodiment that is mutually exclusive with other embodiments.

[0119] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a method for predicting the effect of risk control rules based on big data, including the following steps:

[0120] S1: Collect transaction detail datasets, user activity record datasets, insurance claim datasets, geographical location datasets, and time stamp information datasets from multiple data sources, and preprocess the collected information to form datasets;

[0121] Set the original data sets of the transaction detail datasets, user activity record datasets, insurance claim datasets, geographical location datasets, and time stamp information datasets as D, and the expression is:

[0122]

[0123] where T, U, M, G, and Tm respectively represent the transaction detail datasets, user activity record datasets, insurance claim datasets, geographical location datasets, and time stamp information datasets;

[0124] Each dataset contains its own data points, and the expression is:

[0125]

[0126] where X i represents the i-th dataset, and each dataset contains multiple data points X ij , i represents the data type, and j represents the data point number;

[0127] For the transaction detail datasets, convert the transaction details, and the expression is:

[0128] f(T) = log(1 + T a );

[0129] where f(T) represents the conversion of the transaction detail dataset T, log represents the natural logarithm function, and T a represents the transaction amount;

[0130] For the user activity record datasets, perform preprocessing, and the expression is:

[0131]

[0132] Among them, f(U) represents the value of the normalized user activity record dataset U, min(U) represents the minimum value in the dataset, and max(U) represents the maximum value in the dataset;

[0133] For the insurance claim dataset, obtain its periodic change pattern, and the expression is:

[0134] g(M) = sin(ω M ·t + ω M );

[0135] Among them, g(M) represents the influence of market dynamics changing over time, sin represents the sine function, w M represents the angular frequency of the insurance claim dataset, t represents the time point, represents the phase shift of the insurance claim dataset;

[0136] For the geographical location dataset, perform preprocessing, and the expression is:

[0137] h(G) = exp(-α·d 2 );

[0138] Among them, h(G) represents the influence of the trading location on the trading risk, exp represents the exponential function, α represents the attenuation coefficient, which is a positive constant representing the attenuation speed of the exponential function, and d represents the Euclidean distance between two points;

[0139] For the time - marked information dataset, perform preprocessing, and the expression is:

[0140]

[0141] Among them, k(Tm) represents the influence of the event over time, e represents the base of the natural logarithm, β represents the steepness parameter, a larger β value makes the curve steeper, indicating that the transition interval of the function value from close to 0 to close to 1 becomes narrower, t represents the time point, and τ represents the time threshold;

[0142] Fuse the pre - processed data to form a dataset, and the expression is:

[0143]

[0144] Among them, x ′ represents the pre - processed dataset;

[0145] The said multi - source data includes, but is not limited to, bank transaction systems, social media platforms, financial market APIs, and GPS positioning devices;

[0146] The said transaction details dataset contains the amount, timestamp, and transaction type of each transaction;

[0147] The user activity record data set includes user logins, browsing history, and purchase behavior;

[0148] The insurance claim data set includes the claim application time, claim amount, claim reason, claim processing time, and result;

[0149] The geographical location data set includes the geographical coordinates where the transaction occurred;

[0150] The time stamp information data set contains the date, time, and time zone;

[0151] S2 uses Generative Adversarial Networks (GANs) to construct an anomaly detection model to identify abnormal behaviors in transaction patterns;

[0152] Set the generator network as G, the discriminator network as DL, X as the real transaction data, Z as the random noise vector. The goal of the generator G is to map the noise z into synthetic data of real transaction data, while the goal of the discriminator D is to distinguish between real transaction data x and the synthetic data generated by G;

[0153] In the generator and discriminator networks of GANs, at the same time, the risk control rule parameters are used as part of the input features. The rule parameters include thresholds, rule types, and the number of historical violations. Use one-hot encoding to convert these parameters into numerical forms, and then input them into the anomaly detection model together with the transaction data;

[0154] Use the loss function to measure the difference between real transaction data and synthetic data. The expression is:

[0155]

[0156] Among them, L GAN (G, DL) represents the loss function of the generative adversarial network, P d represents the distribution of real data, P n represents the distribution of noise, represents the average of the variable x under real data, represents the average of the variable z under the noise distribution, log(DL(x)) represents the output of the discriminator DL for real data x converted by the natural logarithm function. The output of the discriminator DL is a probability value between 0 and 1, indicating the probability that the input x is real data;

[0157] Set the anomaly scoring function to identify abnormal behaviors in transaction patterns. The expression is:

[0158]

[0159] Among them, S a represents the anomaly scoring function, which is used to quantify the transaction data point x′ The degree of abnormality, x ′ represents the preprocessed transaction record. α is a constant. When α is larger, the function is more sensitive to the change of DL(x) - μ. This means that the scoring function will change rapidly when DL(x ′ ) is close to μ, resulting in a steep change in the anomaly score. μ represents the average value output by the discriminator DL for the real transaction data during training, and e represents the base of the natural logarithm;

[0160] Set the threshold IO;

[0161] Anomaly scoring function S a The value range is from 0 to 1. When 0 ≤ S a < IO, it means that the transaction pattern is highly consistent with the normal behavior. When IO ≤ S a ≤ 1, it means that the transaction pattern significantly deviates from the normal behavior and is an abnormal behavior;

[0162] The risk control rule parameters include rule types and rule thresholds. The rule types include amount rules and rating rules, and the rule thresholds include that the amount of a single transaction exceeds 5000 yuan and the number of transactions within one day exceeds 10 times;

[0163] S3 constructs a risk control prediction model based on the spatio-temporal graph convolutional neural network STGCN. Input the abnormal behavior data combined with the risk control rule parameters into the risk control prediction model to obtain the predicted risk probability;

[0164] Based on the preprocessed dataset x ′ Construct a spatio-temporal graph. Extract transaction details, user activities, market dynamics, geographical locations, and time stamp information from D ′ and combine them with the risk control rule parameters to form the features of the nodes, that is, the positions and attributes of the nodes in the spatio-temporal graph. Each node v represents a specific spatio-temporal event. Combine all the node sets to form the node set V. Based on the node set and the edge set, generate a spatio-temporal graph. The expression is:

[0165] G st =(V, E);

[0166] Among them, G st represents the spatio-temporal graph, s represents the geographical coordinates, t represents the time stamp, V represents the node set in the spatio-temporal graph, and E represents the edge set in the spatio-temporal graph;

[0167] Set the spatio-temporal adjacency matrix as A. The spatio-temporal adjacency matrix A is a two-dimensional matrix, where the rows and columns respectively correspond to each node in the graph. If there is a direct connection between i and j, then the element a ij in A will be a non-zero value, indicating their degree of association; if there is no direct connection, then a ij has a value of 0;

[0168] Calculate the element aij in the adjacency matrix using the Gaussian kernel function K, and the expression is:

[0169]

[0170] where a ij represents the element of the adjacency matrix between node i and node j, α is the time decay coefficient, which controls the rate of influence on the adjacency strength a t ij As the value of α increases, the adjacency strength will decrease rapidly with the increase of the time distance; conversely, when the value of α is small, the rate of change of the adjacency strength with the time distance is slow. d ij t ij represents the distance between node i and node j in the time dimension, and β represents the spatial decay coefficient, which controls the influence on the adjacency strength a s ij ij d s ij represents the distance between node i and node j in the spatial dimension;

[0171] In STGCN, a spatio-temporal graph convolutional layer is used to propagate information. The expression for the new hidden state after being processed by the spatio-temporal graph convolutional layer is:

[0172]

[0173] where h v new represents the new hidden state of node v after being processed by the spatio-temporal graph convolutional layer, σ is the activation function, W s represents the spatial weight matrix, a vu represents the adjacency weight between node v and node u, h u represents the current hidden state of node u, W t represents the time weight matrix, h v represents the current hidden state of node v;

[0174] Set the original feature vector of each node v in the spatio-temporal graph to x v ,

[0175] Attach the anomaly score S a to the feature vector x v to form an enhanced feature vector, and the expression is:

[0176]

[0177] where x v avg represents the enhanced feature vector;​​

[0178] Input all enhanced feature vectors into the risk control prediction model for training. For each node, the enhanced feature vector x v avg Update its hidden state. After processing through multiple layers of STGCN, obtain the final hidden state h of each node v f ;

[0179] Pass the hidden state to the fully connected layer to predict the risk probability. The output layer expression is:

[0180]

[0181] where p r represents the predicted risk probability, σ represents the activation function, and F represents the fully connected layer;

[0182] Set the threshold Υ;

[0183] The value range of the predicted risk probability p r is from 0 to 1. When the value of p r is less than Υ, it represents low risk. When the value of p r is greater than or equal to Υ, it represents high risk.

[0184] S4 Based on abnormal behavior, combined with insurance claim assessment, fraud detection, and insurance market fluctuations, comprehensively evaluate the transaction behavior to obtain a comprehensive evaluation result;

[0185] Collect historical claim data from historical data, including but not limited to claim application time, claim amount, claim reason, claim processing time, and result;

[0186] Preprocess the claim historical data, including data cleaning, missing value handling, outlier detection, feature selection, and engineering;

[0187] Use the deep learning model LSTM to build a claim settlement habit prediction model P ins using the historical claim data to input into the claim settlement habit prediction model for training;

[0188] Use the cross-validation method to verify the output score of the model and tune the parameters to improve the generalization ability and prediction accuracy of the model;

[0189] Based on the output score of the claim settlement habit prediction model, calculate the claim habit score. The expression is:

[0190] I s = f(P ins (Ru),We Ins );

[0191] where Is It represents the claim settlement habit score, indicating the degree of change in the claim settlement and loss assessment habit predicted according to risk control rules. f represents the claim function, which is used to calculate the claim settlement habit score and is composed of weighted average and linear combination. Ru represents the risk control rule parameter, and We ins represents the weight vector of the claim settlement habit score;

[0192] When it is predicted that the claim frequency will increase and the average claim amount will also rise, the claim settlement habit score will increase accordingly, and vice versa;

[0193] Based on the user's historical credit performance and payment behavior, calculate the insurance claim assessment score. The expression is:

[0194]

[0195] Among them, C s represents the insurance claim assessment score, and its value range is between 0 and 1. A value close to 0 indicates a low credit risk, and a value close to 1 indicates a high credit risk. H represents the user's historical credit score, and μ H represents the historical average credit score, D represents the average payment delay days, e represents the base of the natural logarithm, and w1 and w2 are the weight coefficients of the user's historical credit score and average payment delay days respectively;

[0196] Based on the anomaly detection model, calculate the fraud detection score. The expression is:

[0197]

[0198] Among them, F d represents the fraud detection score, and its value range is between 0 and 1. S min represents the minimum value of the anomaly score, and S max represents the maximum value of the anomaly score;

[0199] Based on the periodic change pattern in the insurance claim dataset M, calculate the insurance market volatility score. The expression is:

[0200]

[0201] Among them, σ represents the volatility index of the stock price, and its value range is between 0 and 1. The larger the value of the index, the higher the volatility of the insurance market and the greater the uncertainty. σ min and σ max represent the minimum and maximum values of the volatility index respectively,

[0202] Input the claim settlement habit score together with the insurance claim assessment score, fraud detection score, and insurance market volatility score into the calculation formula of the comprehensive evaluation result. The expression is:

[0203]

[0204] Among them, R c represents the comprehensive evaluation result, and W c , W f , W m , W i respectively represent the weights of the insurance claim assessment score, fraud detection score, insurance market volatility score, and claim settlement habit score. λ1, λ2, λ3, and λ4 represent adjustment factors used to control the degree of non-linear influence of each score on the comprehensive evaluation result R c when it deviates from its benchmark value, and μ C , μ F , μ M , μ I are the average values of the insurance claim assessment score, fraud detection score, insurance market volatility score, and claim settlement habit score, respectively.

[0205] S5 formulates corresponding risk control and prevention strategies based on the predicted risk probability and the comprehensive evaluation result;

[0206] Set the market sentiment indicator as M e , which is obtained by analyzing the sentiment tendencies of social media, news reports, and investor forums, and its value range is between 0 and 1. 0 indicates extremely pessimistic market sentiment, and 1 indicates extremely optimistic market sentiment;

[0207] Combining the predicted risk probability p r , the comprehensive evaluation result R c and the market sentiment indicator M e to construct a decision function, the expression of which is:

[0208]

[0209] Among them, D f represents the decision function score, γ represents the market sentiment adjustment factor used to adjust the intensity of the influence of the market sentiment M e on the final decision score D f . When the value of γ increases, the influence of the market sentiment on the decision will also increase. δ represents the non-linear influence index of the predicted risk probability p r , which determines the degree of non-linear influence of p r on the output of the decision function D f . When δ is greater than 1, as the value of p r increases, its contribution to D f also increases rapidly. η represents the non-linear influence index of the comprehensive evaluation result R c , which determines the degree of non-linear influence of R c on the output of the decision function D f . When η is greater than 1, as R cAs its value increases, its contribution to D f also accelerates. ξ represents the balance coefficient, which is used to prevent the predicted risk probability p r and the comprehensive evaluation result R c from causing the decision score D f to increase infinitely in the case of being particularly high. By adjusting ξ, the upper limit of D f can be controlled to ensure that the decision score does not exceed a reasonable range;

[0210] The value range of the decision function score D f is from 0 to 1;

[0211] When D f is close to 0, it indicates that the trading risk is low, and fewer risk control measures can be taken;

[0212] When D f is close to 1, it indicates that the trading risk is high, and strict risk control measures are required;

[0213] Set a low-risk threshold DF and a high-risk threshold GF;

[0214] When D f <DF, it means that the trading risk is low, and continue to monitor without taking action;

[0215] When DF ≤ D f <GF, it means that the trading risk is medium, start the early warning mechanism, strengthen the review of relevant accounts, and require users to provide verification information of mobile phone verification codes;

[0216] When D f ≥ GF, it means that the trading risk is extremely high, immediately freeze the transaction and contact the user to confirm the authenticity of the transaction. At the same time, report the suspicious activity to the regulatory agency;

[0217] When the amount of a single transaction exceeds 5000 yuan or the number of transactions within a day exceeds 10 times, immediately freeze the transaction and contact the user to confirm the authenticity of the transaction. At the same time, report the suspicious activity to the regulatory agency;

[0218] The low-risk threshold DF is usually 0.3, and the high-risk threshold GF is 0.7.

[0219] S6 inputs real-time data into the model to obtain real-time risk control prevention strategies and execute the strategies;

[0220] Collect real-time data from multiple data sources, including real-time transaction details, real-time user activities, real-time market dynamics, real-time geographical location information, and real-time time stamp information;

[0221] Preprocess the real-time data respectively, and then input it into the anomaly detection model to obtain real-time anomaly scores;

[0222] Combine the real-time anomaly score with the risk control rule parameters, input them into the risk control prediction model for real-time risk assessment, and output the real-time predicted risk probability;

[0223] Calculate the real-time comprehensive evaluation result based on the real-time anomaly score;

[0224] Input the real-time predicted risk probability and the real-time comprehensive evaluation result into the decision function formula to obtain the real-time decision function score, and at the same time execute the corresponding risk control strategy.

[0225] This embodiment also provides a risk control rule effect prediction system based on big data, including:

[0226] The acquisition module is responsible for collecting transaction details, user activities, market dynamics, geographical locations, and time stamp information from multiple data sources, preprocessing these data, and converting them into a format that can be used for model analysis;

[0227] The detection and scoring module is responsible for using the generative adversarial network GANs to identify abnormal behaviors in transaction patterns, assign an anomaly score to each transaction data point, and quantify the degree of deviation from normal behavior;

[0228] The risk prediction module predicts the risk probability of a single transaction based on the spatio-temporal graph convolutional neural network STGCN, and conducts a comprehensive evaluation by combining factors such as anomaly scores, insurance claim assessments, fraud detections, and insurance market fluctuations;

[0229] The decision-making module is responsible for combining the predicted risk probability, the comprehensive evaluation result, and the market sentiment index to generate a decision function score, and formulating corresponding risk control strategies according to the decision function score;

[0230] The execution module is responsible for receiving and processing new data in real time, dynamically evaluating risks, and immediately executing the corresponding risk control strategies.

[0231] This embodiment also provides a computer device applicable to the case of the risk control rule effect prediction method based on big data, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the risk control rule effect prediction method based on big data as proposed in the above embodiment.

[0232] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.

[0233] The present embodiment also provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the risk control rule effect prediction method based on big data as proposed in the above embodiment; the storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0234] In summary, the present invention ensures the immediacy and accuracy of risk control strategies by: inputting real-time data; enhancing the interpretability of model decisions and improving customers’ understanding and trust in the claims process by optimizing model structure and algorithms; and the risk control rule effect prediction model provides a scientific basis for rule adjustment, enabling insurance companies to make decisions based on quantitative analysis, thereby improving the efficiency and effectiveness of rule adjustment.

[0235] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the advancement of the present invention, experimental simulation data of the risk control rule effect prediction method based on big data is provided.

[0236] In this embodiment, a financial transaction scenario involving multiple data types is simulated, aiming to evaluate and optimize a new intelligent financial risk control system. The experimental data is sourced from a virtual bank transaction system, a social media platform, a financial market API, and a GPS positioning device. These data are collected and preprocessed to form five datasets: transaction details, user activities, market dynamics, geographical locations, and time-stamped information.

[0237] The preprocessing of the datasets includes logarithmic transformation of transaction amounts, normalization of user activity records, extraction of periodic change patterns in market dynamics, exponential decay processing of geographical locations, and Sigmoid function transformation of time-stamped information. The preprocessed data is fused to form a comprehensive dataset.

[0238] Next, a generative adversarial network (GANs) is used to build an anomaly detection model, which can identify abnormal behaviors in transaction patterns. By setting an anomaly scoring function and a threshold, normal and abnormal transactions can be effectively distinguished.

[0239] Subsequently, a spatio-temporal graph convolutional neural network (STGCN) is used to build a risk control prediction model. This model takes abnormal behavior data as input and outputs the predicted risk probability. This stage involves constructing a spatio-temporal graph, calculating the spatio-temporal adjacency matrix, and using spatio-temporal graph convolutional layers to propagate information.

[0240] Finally, the abnormal behaviors are comprehensively evaluated in combination with insurance claim assessment, fraud detection, and insurance market fluctuations to obtain a comprehensive evaluation result. Based on the predicted risk probability and the comprehensive evaluation result, corresponding risk control prevention strategies are formulated.

[0241] Specifically, as shown in Table 1:

[0242] Table 1 Experimental Record Table

[0243]

[0244]

[0245] By observing the table data, the decision function scores (D f ) under different transaction cases can be seen. The decision function score D f reflects the level of transaction risk. The closer its value is to 1, the higher the transaction risk.

[0246] In the first row, the transaction amount is small (2000), and both the user activity record and the market dynamics are at normal levels. Therefore, the decision function score D f is low (0.3), indicating a low transaction risk.

[0247] In contrast, the transaction amount in the fourth row is relatively large (10,000). Although the geographical location risk is relatively low, due to the combined effect of other factors (such as user activities and market dynamics), the decision function score D f is as high as 0.9, indicating that this is a high-risk transaction.

[0248] By comparing the decision function scores of different cases, it can be clearly seen that the beneficial effect of the present invention lies in its ability to effectively identify and quantify potential risks in transactions, thereby providing accurate risk assessments and corresponding risk control strategy recommendations for financial institutions. Compared with traditional methods, this method more comprehensively considers the impact of multiple data sources, improving the accuracy and timeliness of risk detection.

[0249] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A risk control rule effect prediction method based on big data, characterized by: include, Collect transaction details data sets, user activity record data sets, insurance claims data sets, geographic location data sets, and time stamp information data sets from multiple data sources, and pre-process the collected information to form a data set; Use generative adversarial networks (GANs) to build an anomaly detection model to identify abnormal behaviors in trading patterns; A risk control prediction model is built based on the spatiotemporal graph convolutional neural network STGCN. The abnormal behavior data is combined with the risk control rule parameters and input into the risk control prediction model to obtain the predicted risk probability. Based on abnormal behavior combined with insurance claims assessment, fraud detection, and insurance market fluctuations, a comprehensive assessment of transaction behavior is conducted to obtain a comprehensive assessment result; Formulate corresponding risk control and prevention strategies based on the predicted risk probability and comprehensive assessment results; Input real-time data into the model, obtain real-time risk control and prevention strategies, and execute the strategies.

2. The risk control rule effect prediction method based on big data according to claim 1, characterized in that: The specific steps of collecting transaction details data set, user activity record data set, insurance claim data set, geographic location data set and time stamp information data set from multiple data sources and preprocessing the collected information to form a data set are as follows: Assume that the original data set of transaction details dataset, user activity record dataset, insurance claim dataset, geographic location dataset, and time stamp information dataset is D, and the expression is: Among them, T, U, M, G, and Tm represent the transaction details dataset, user activity record dataset, insurance claim dataset, geographic location dataset, and time stamp information dataset, respectively; Each data set contains its own data points, expressed as: Among them, X i Represents the i-th data set, each of which contains multiple data points X ij , i represents the data type, j represents the data point number; For the transaction details data set, the transaction details data is converted into the following expression: f(T) = log(1 + T); Where f(T) represents the transformation of the transaction details dataset T, log represents the natural logarithm function, and T a Indicates the transaction amount; Preprocess the user activity record dataset, and the expression is: Where f(U) represents the value of the normalized user activity record dataset U, min(U) represents the minimum value in the dataset, and max(U) represents the maximum value in the dataset; For the insurance claims data set, obtain its periodic change pattern, the expression is: g(M)=sin(ω M ·t+φ M ); Where g(M) represents the impact of market dynamics over time, sin represents the sine function, and w M represents the angular frequency of the insurance claim data set, t represents the time point, represents the phase shift of the insurance claims dataset; For the geographic location dataset, preprocessing is performed, and the expression is: h(G)=exp(-α·d 2 ); Among them, h(G) represents the impact of the transaction location on the transaction risk, exp represents the exponential function, α represents the attenuation coefficient, and d represents the Euclidean distance between two points; Preprocessing is performed on the time tag information dataset, and the expression is: Among them, k(Tm) represents the impact of the event over time, e represents the base of the natural logarithm, β represents the steepness parameter, t represents the time point, and τ represents the time threshold; The preprocessed data is fused to form a data set, which is expressed as: Among them, D ′ Represents the preprocessed dataset.

3. The risk control rule effect prediction method based on big data according to claim 2, characterized in that: The anomaly detection model is constructed by using the generative adversarial network GANs to identify abnormal behaviors in transaction patterns. The specific steps are as follows: Set the generator network to G, the discriminator network to DL, X to real transaction data, and Z to random noise vector; In the generator and discriminator networks of GANs, the risk control rule parameters are taken as part of the input features, converted into numerical form using one-hot encoding, and then input into the anomaly detection model together with the transaction data; The loss function is used to measure the difference between real transaction data and synthetic data. The expression is: Among them, L GAN (G, DL) represents the loss function of the generated adversarial network, P d represents the distribution of real data, P n represents the distribution of noise, represents the average of variable x under real data, represents the average of the variable z under the noise distribution, log(DL(x)) represents the output of the discriminator DL ​​for the real data x transformed by the natural logarithmic function; Set the anomaly scoring function to identify abnormal behaviors in trading patterns. The expression is: Among them, S a represents the anomaly scoring function used to quantify the transaction data point x ′ The degree of abnormality, x ′ represents the preprocessed transaction records, α is a constant, μ represents the average value of the discriminator DL ​​output on the real transaction data during the training process, and e represents the base of the natural logarithm; Set threshold IO; Abnormal scoring function S a has a value range of 0 to 1. When 0 ≤ S a <IO, it indicates that the trading pattern is highly consistent with normal behavior. When IO ≤ S a ≤ 1, it indicates that the trading pattern significantly deviates from normal behavior and is an abnormal behavior.

4. The risk control rule effect prediction method based on big data according to claim 3, characterized in that: The risk control prediction model is constructed based on the spatiotemporal graph convolutional neural network STGCN, and the abnormal behavior data is combined with the risk control rule parameters into the risk control prediction model to obtain the predicted risk probability. The specific steps are as follows: Based on the preprocessed dataset x ′ Construct a space-time graph, starting from D ′ The transaction details, user activities, market dynamics, geographic location, and time stamp information are extracted and combined with the risk control rule parameters to form the characteristics of the node, that is, the position and attributes of the node in the space-time graph. Each node v represents a specific space-time event. All nodes are collected to form a node set V. Based on the node set and edge set, the space-time graph is generated. The expression is: G st =(V,E); Among them, G st represents a space-time graph, s represents geographic coordinates, t represents a timestamp, V represents a node set in the space-time graph, and E represents an edge set in the space-time graph; Let the spatiotemporal adjacency matrix be A. The spatiotemporal adjacency matrix A is a two-dimensional matrix, where the rows and columns correspond to each node in the graph. If there is a direct connection between i and j, then the element a in A ij will be a non-zero value, indicating their degree of association; if there is no direct connection, then a ij The value of is 0; Use the Gaussian kernel function K to calculate the middle element aij of the adjacency matrix, the expression is: Among them, a ij represents the adjacency matrix element between node i and node j, α is the time decay coefficient, d t ij represents the distance between node i and node j in the time dimension, β represents the spatial attenuation coefficient, and d s ij Represents the distance between node i and node j in the spatial dimension; In STGCN, the spatiotemporal graph convolution layer is used to propagate information. The new hidden state after processing by the spatiotemporal graph convolution layer is expressed as: Among them, h v new represents the new hidden state of node v after being processed by the spatiotemporal graph convolution layer, σ is the activation function, and W s represents the spatial weight matrix, a vu represents the adjacency weight between node v and node u, h u represents the current hidden state of node u, W t represents the time weight matrix, h v Represents the current hidden state of node v; Set the original feature vector of each node v in the space-time graph to x v , The abnormal score S a Append to the feature vector x v On the basis of, an enhanced feature vector is formed, which is expressed as: Among them, x v avg represents the enhanced feature vector; All enhanced feature vectors are input into the risk control prediction model to complete the training, and the enhanced feature vector x of each node is v avg Update its hidden state, and after processing through multiple layers of STGCN, obtain the final hidden state h of each node v f ; The hidden state is passed to the fully connected layer to predict the risk probability, and the output layer expression is: Among them, p r represents the predicted risk probability, σ represents the activation function, and F represents the fully connected layer; Set a threshold value Y; Predicted risk probability p r The value range is from 0 to 1. r When the value is less than Υ, it indicates low risk. r When the value of is greater than or equal to Υ, it indicates high risk.

5. The risk control rule effect prediction method based on big data according to claim 4, characterized in that: The above-mentioned process of conducting a comprehensive assessment of transaction behaviors based on abnormal behaviors combined with insurance claim assessment, fraud detection, and insurance market fluctuations to obtain a comprehensive assessment result includes the following specific steps: Collect historical claims data from historical data; Pre-process historical claims data; The deep learning model LSTM is used to build a claim settlement habit prediction model P based on historical claim data. ins , use historical claims data to input the claims loss assessment habit prediction model to complete the training; Use cross-validation to verify the model's output scores and tune parameters to improve the model's generalization ability and prediction accuracy; Based on the output score of the claim settlement habit prediction model, the claim settlement habit score is calculated, and the expression is: I s =f(P ins (Ru),We Ins ); Among them, I s is the claim settlement habit score, f represents the claim settlement function, Ru represents the risk control rule parameter, and We ins The weight vector representing the claim habit score; Based on the user's historical credit performance and payment behavior, the insurance claim assessment score is calculated as follows: Among them, C s represents the insurance claim assessment score, H represents the user's historical credit score, μ H represents the historical average credit score, D represents the average number of days of delayed payment, e represents the base of the natural logarithm, w1 and w2 are the weight coefficients of the user's historical credit score and the average number of days of delayed payment, respectively; Based on the anomaly detection model, the fraud detection score is calculated as follows: Among them, F d represents the fraud detection score, S min represents the minimum value of the anomaly score, S max represents the maximum value of the anomaly score; The insurance market volatility score is calculated based on the periodic change pattern in the insurance claims dataset M, and the expression is: Among them, σ represents the volatility index of stock prices, σ min and σ max Represent the minimum and maximum values ​​of the volatility index, The claim habit score, insurance claim assessment score, fraud detection score, and insurance market volatility score are input into the calculation formula of the comprehensive evaluation result, and the expression is: Among them, R c represents the comprehensive evaluation result, W c , W f , W m , W i They represent the weights of the insurance claim assessment score, fraud detection score, insurance market volatility score, and claim habit score, respectively. λ1, λ2, λ3, and λ4 represent adjustment factors, which are used to control the impact of each score deviating from its baseline value on the comprehensive evaluation result R. c The nonlinear influence degree, μ C , μ F , μ M , μ I They are the averages of the insurance claims assessment score, fraud detection score, insurance market volatility score, and claims habits score.

6. The risk control rule effect prediction method based on big data according to claim 5, characterized in that: Based on the predicted risk probability and comprehensive assessment results, the corresponding risk control and prevention strategies are formulated. The specific steps are as follows: Set the market sentiment index to M e ; Combined with the predicted risk probability p r , Comprehensive evaluation results R c and market sentiment index M e Construct a decision function, the expression is: Among them, D f represents the decision function score, γ represents the market sentiment adjustment factor, and δ represents the predicted risk probability p r The nonlinear influence index, η represents the comprehensive evaluation result R c The nonlinear influence index, ξ represents the balance coefficient; Decision function score D f The value range of is 0 to 1; When D f When it is close to 0, it indicates that the transaction risk is low and fewer risk control measures can be taken; When D f When it is close to 1, it indicates that the transaction risk is high and strict risk control measures need to be taken.

7. The risk control rule effect prediction method based on big data according to claim 6, characterized in that: The real-time data is input into the model to obtain the real-time risk control and prevention strategy, and the strategy is executed. The specific steps are as follows: Collect real-time data from multiple data sources; Preprocess the real-time data separately and then input it into the anomaly detection model to obtain the real-time anomaly score; Combine the real-time anomaly score with the comprehensive assessment results, input them into the risk control prediction model for real-time risk assessment, and output the real-time predicted risk probability; Obtain real-time comprehensive evaluation results based on real-time anomaly score calculation; Based on the real-time predicted risk probability and real-time comprehensive evaluation results, the decision function formula is input to obtain the real-time decision function score, and the corresponding risk control strategy is executed at the same time.

8. A risk control rule effect prediction system based on big data, based on the risk control rule effect prediction method based on big data according to any one of claims 1 to 7, characterized in that: include: The acquisition module is responsible for collecting transaction details, user activities, market dynamics, geographic location and time stamp information from multiple data sources, and preprocessing this data to convert it into a format that can be used for model analysis; The detection and scoring module is responsible for using generative adversarial networks (GANs) to identify abnormal behaviors in trading patterns and assign an anomaly score to each trading data point to quantify its degree of deviation from normal behavior; The risk prediction module predicts the risk probability of a single transaction based on the spatiotemporal graph convolutional neural network STGCN, and conducts a comprehensive assessment based on anomaly scoring, insurance claim assessment, fraud detection, and insurance market volatility factors; The decision-making module is responsible for combining the predicted risk probability, comprehensive assessment results and market sentiment indicators to generate a decision function score and formulate corresponding risk control strategies based on the decision function score; The execution module is responsible for receiving and processing new data in real time, dynamically assessing risks, and immediately executing corresponding risk control strategies.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the risk control rule effect prediction method based on big data described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the risk control rule effect prediction method based on big data described in any one of claims 1 to 7 are implemented.

Images