A method and system for predicting debt evasion behavior after overdue

Abstract

The present invention relates to the fields of artificial intelligence and financial risk control technology, and in particular to a method and system for predicting post-debt default behavior. The method collects multidimensional user behavior data, performs data cleaning, standardization, and feature fusion to construct time series data. It also uses a bidirectional long-short-term memory network combined with deep reinforcement learning to dynamically model user behavior and output risk prediction data. A comprehensive risk score is generated based on a multi-objective decision-making algorithm and fuzzy logic. After verification with an auxiliary statistical model, the model is updated using incremental learning and Bayesian optimization to ultimately generate accurate prediction results. This method significantly improves the accuracy and real-time nature of predicting post-debt default behavior, providing a scientific basis for financial risk control.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and financial risk control technology, and in particular to a method and system for predicting debt evasion behavior after overdue payments. Background Art

[0002] Technology for predicting post-debt default behavior aims to proactively identify potential default risks through in-depth analysis of user behavior after a default, providing financial institutions with a basis for proactive risk management. Existing technologies primarily rely on social network data and construct XGBoost-based classification models to identify default groups. While this method achieves some degree of risk identification, it suffers from limitations such as insufficient data real-time availability, delayed model updates, a strong reliance on manual labeling, and limited feature extraction. Furthermore, traditional methods are limited in capturing the temporal dynamics of user behavior, making the timeliness and accuracy of risk predictions difficult to meet financial risk control requirements. Summary of the Invention

[0003] In response to the many problems existing in the above-mentioned prior art, the present invention provides a method and system for predicting debt evasion after overdue. The present invention collects multi-dimensional behavioral data of users, extracts features after data cleaning and standardization, and fuses them into time series data through a fixed window method; then, a bidirectional long short-term memory network is used to model the time series data to capture the bidirectional time series dependency of user behavior, and deep reinforcement learning technology is combined to dynamically adjust the model parameters to reduce prediction errors; finally, a comprehensive risk score and numerical conversion are performed on the prediction results through a multi-objective decision algorithm and fuzzy logic, and cross-validation is performed using auxiliary statistical methods. Finally, the adaptive update of the model is achieved with the help of incremental learning and Bayesian optimization, thereby generating accurate final prediction results. The present invention realizes real-time prediction and early warning of debt evasion after overdue, providing financial institutions with an efficient risk prevention and control tool.

[0004] A method for predicting debt default behavior after overdue payment, comprising the following steps:

[0005] Collect multi-dimensional behavioral data of users, perform data cleaning and data standardization on the behavioral data, extract features from the standardized data, and generate a fused feature data set;

[0006] Constructing time series data based on the fused feature dataset, using a bidirectional long short-term memory network for training to generate time series risk prediction data for each user, and using deep reinforcement learning to adjust parameters of the bidirectional long short-term memory network to form a risk prediction model and output risk prediction data;

[0007] Based on the risk prediction data, a multi-objective decision-making algorithm is used to calculate a comprehensive risk score, and fuzzy logic is used to numerically adjust the comprehensive risk score to form decision intervention data. At the same time, an auxiliary statistical model is used to cross-validate the risk prediction data, and statistically validated risk assessment data is output;

[0008] Based on the statistically verified risk assessment data as feedback, the risk prediction model is updated with parameters using an incremental learning method, and the hyperparameters of the risk prediction model are adjusted using an adaptive optimization algorithm to generate a final prediction result.

[0009] Preferably, the multi-dimensional behavioral data is collected from user account transaction records, user social interaction records and user device usage records in a preset data format, and the data beyond the statistical range is eliminated through an outlier detection algorithm. At the same time, a missing data filling algorithm is used to compensate for missing items in the data, and the collected data is normalized using a minimum and maximum normalization method, thereby generating standardized original data with a unified numerical range.

[0010] Preferably, the features in the normalized raw data are calculated using a fixed-window-based feature extraction algorithm. The calculation process quantifies the user's account changes, social interaction frequency, and device usage patterns respectively, and accumulates the quantified results according to a preset formula to form a fused feature data set.

[0011] Preferably, the time series data constructed based on the fusion feature data set is trained using a bidirectional long short-term memory network, which sets a forward layer and a backward layer in the network structure and uses a backpropagation algorithm to update the network parameters to generate time series risk prediction data for each user.

[0012] Preferably, the deep reinforcement learning adopts a proximal policy optimization algorithm to dynamically adjust the weights of the bidirectional long short-term memory network according to a preset reward function, wherein the reward function uses the risk prediction error as feedback input, thereby forming a risk prediction model and outputting risk prediction data.

[0013] Preferably, the multi-objective decision-making algorithm calculates the respective risk scores for financial behavior risk, social behavior risk and device behavior risk according to preset weighting coefficients, and generates comprehensive risk score data through accumulation operation; the comprehensive risk score data is converted through a preset membership function, and the continuous risk score is discretized into decision parameters to form decision intervention data.

[0014] Preferably, the auxiliary statistical model adopts a five-fold cross-validation method to divide the risk prediction data into a training set and a validation set, and compares the consistency between the risk prediction data and the decision intervention data by calculating statistical indicators, thereby outputting statistically verified risk assessment data.

[0015] Preferably, the incremental learning method adopts an online gradient descent algorithm. When the newly collected user behavior data reaches a preset threshold, the new data is integrated into the risk prediction model and the model parameters are updated in real time.

[0016] Preferably, the adaptive optimization algorithm adopts Bayesian optimization technology to automatically adjust the hyperparameters of the risk prediction model within a preset hyperparameter search range, and generates a final prediction result based on statistically verified risk assessment data as feedback.

[0017] A system for predicting overdue debt evasion behavior, used to implement the method for predicting overdue debt evasion behavior, comprising:

[0018] The collection module is used to collect multi-dimensional behavioral data such as user account transaction records, user social interaction records, and user device usage records according to a preset data collection plan;

[0019] A data processing module is used to perform data cleaning, abnormal data removal, missing data filling, and minimum and maximum normalization on the multi-dimensional behavior data, and extract features from the normalized data to generate a fused feature data set;

[0020] A time series modeling module is used to construct time series data based on the fused feature dataset, train the time series data using a bidirectional long short-term memory network, generate time series risk prediction data for each user, and adjust the weight parameters of the bidirectional long short-term memory network using deep reinforcement learning to form a risk prediction model and output risk prediction data;

[0021] A decision-making module is configured to calculate risk scores for each risk dimension using a multi-objective decision-making algorithm based on the risk prediction data, calculate a comprehensive risk score according to a preset weighting coefficient, perform numerical conversion on the comprehensive risk score using a preset membership function to form decision intervention data, and simultaneously use an auxiliary statistical module to cross-validate the risk prediction data and output statistically validated risk assessment data;

[0022] The model updating module is used to update the parameters of the risk prediction model based on the risk assessment data feedback using the online gradient descent incremental learning method, and to automatically adjust the hyperparameters of the risk prediction model within a preset search interval using the Bayesian optimization technology to generate the final prediction result.

[0023] Compared with the prior art, the advantages and beneficial effects of the present invention are:

[0024] The present invention uses multimodal data fusion technology to achieve comprehensive collection and standardized processing of multi-dimensional user behavior data, thereby generating a high-quality fusion feature data set;

[0025] This invention uses a bidirectional long short-term memory network and deep reinforcement learning technology to achieve temporal dynamic modeling and real-time parameter adjustment of users' overdue behavior, thereby outputting stable risk prediction data.

[0026] This invention uses a multi-objective decision-making algorithm and fuzzy logic technology to achieve quantitative integration of different risk dimensions and the generation of decision-making intervention data;

[0027] The present invention realizes real-time updating and adaptive optimization of the risk prediction model through cross-validation of auxiliary statistical models and online gradient descent and Bayesian optimization technology, thereby generating the final prediction results.

[0028] The comprehensive application of these technical means not only solves the problems of poor data real-timeness and untimely model updates in existing methods, but also significantly improves the accuracy of risk assessment and the robustness of prediction models, providing scientific and automated technical support for financial risk control. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 Schematic diagram of the process of the present invention;

[0030] Figure 2 Schematic diagram of deep reinforcement learning in the present invention;

[0031] Figure 3 It is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0032] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of well-known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0033] The terms used herein are only for describing specific embodiments and are not intended to limit the present disclosure. The terms "comprise," "include," etc. used herein indicate the presence of the features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0034] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art unless otherwise defined. It should be noted that the terms used herein should be interpreted as having a meaning consistent with the context of this specification and should not be interpreted in an idealized or overly rigid manner.

[0035] like Figure 1 As shown, a method for predicting debt default behavior after overdue payment includes the following steps:

[0036] Collect multi-dimensional behavioral data of users, perform data cleaning and data standardization on the behavioral data, extract features from the standardized data, and generate a fused feature data set;

[0037] Aiming at predicting debt evasion behavior after overdue payments, the present invention first constructs a basic data set reflecting user behavior characteristics through data collection and preprocessing. The overall process includes collecting multi-dimensional behavior data of users, cleaning and standardizing the collected data, and extracting features from the standardized data to generate a fused feature data set.

[0038] During the data collection phase, this step uses a pre-set data collection scheme to obtain behavioral data such as user account transaction records, user social interaction data, and user device usage records from multiple information sources. Subsequently, data cleaning techniques are used to remove erroneous data and detect outliers from the raw data, and a missing data filling algorithm is applied to compensate for missing items in the data. Next, numerical normalization techniques are used to uniformly map data of different dimensions and numerical ranges to a fixed interval, thereby generating standardized raw data. This standardized data serves as input for subsequent feature extraction. Through segmented statistics and calculations, key information reflecting user account changes, social interaction frequency, and device usage patterns is extracted. These indicators are then integrated according to pre-set rules to form a comprehensive fused feature dataset. This dataset not only has unified and standardized data structure but also fully reflects the temporal dynamics of user behavior in terms of content, providing a high-quality input foundation for subsequent risk prediction model training and parameter optimization. Through the above data collection and preprocessing processes, the present invention achieves comprehensive and systematic recording and processing of user behavior at the data level, ensuring the accuracy and stability of the subsequent prediction process while providing sufficient data support and a reliable theoretical basis for risk assessment.

[0039] Preferably, the multi-dimensional behavioral data is collected from user account transaction records, user social interaction records and user device usage records in a preset data format, and the data beyond the statistical range is eliminated through an outlier detection algorithm. At the same time, a missing data filling algorithm is used to compensate for missing items in the data, and the collected data is normalized using a minimum and maximum normalization method, thereby generating standardized original data with a unified numerical range.

[0040] In this invention, the multi-dimensional behavioral data specifically consists of user account transaction records, social interaction records, and device usage records. This data is collected automatically according to a pre-set data format. First, the system extracts user account transaction records, social interaction records, and device usage records from financial trading platforms, social media, and mobile device logs according to a fixed data template. Each data type includes a clear timestamp, numerical indicator, and relevant identifier.

[0041] After collection, all types of data are preliminarily screened using an outlier detection algorithm. The algorithm uses a statistical distribution model (such as a normal distribution or exponential distribution model) to calculate the mean and standard deviation of each data item and eliminates values ​​that exceed the set statistical interval. At the same time, a missing data filling algorithm is used to compensate for missing items in the data based on the mean or median of the same user's historical data to ensure data integrity. Next, the minimum and maximum normalization method is used on the data processed as described above to map each data item to a fixed numerical range (for example, [0,1]). The normalized data is the standardized original data. This process uses programming to achieve automated data processing, ensuring consistency in data format and numerical range across all sources, and providing unified and reliable input data for subsequent feature extraction.

[0042] Preferably, the features in the normalized raw data are calculated using a fixed-window-based feature extraction algorithm. The calculation process quantifies the user's account changes, social interaction frequency, and device usage patterns respectively, and accumulates the quantified results according to a preset formula to form a fused feature data set.

[0043] After generating the standardized raw data, the present invention further uses a fixed window-based feature extraction algorithm to quantify the key information in the data to construct a fused feature data set.

[0044] In specific operations, the system first sets a fixed time window, such as an hourly, daily or weekly window, and divides the standardized raw data into multiple sub-datasets according to the window; in each window, the user account transaction records, social interaction records and device usage records are statistically calculated. For user account transaction records, the system calculates the average value, volatility and transaction frequency of the transaction amount in each window; for user social interaction records, the number of interactions, information transmission frequency and interaction time distribution in each window are counted; for user device usage records, the device activation time and usage frequency are quantified. Each statistical indicator is calculated using a preset formula. For example, the window mean calculation formula for a certain indicator is set as:

[0045]

[0046] Among them, μ w is the mean value in the window, N represents the total number of data in the window, x i is the i-th data value. Subsequently, the system performs weighted accumulation of various statistical indicators according to preset rules to form the final fusion feature data set, where each weight coefficient is predetermined based on historical data performance and expert experience. This weighted accumulation process ensures that the importance of different dimensional data can be reflected when fused, and the accumulation formula is:

[0047]

[0048] Among them, F represents the fusion feature value, α j is the weight coefficient of the jth feature, f j The jth feature within the window is the statistic. Ultimately, the fused feature values ​​generated within all windows are integrated into continuous time series data, providing data support for subsequent risk prediction model training. This fixed-window feature extraction method has the advantages of simple operation, clear parameters, and strong repeatability, and can stably reflect key behavioral characteristics across different user groups.

[0049] Constructing time series data based on the fused feature dataset, using a bidirectional long short-term memory network for training to generate time series risk prediction data for each user, and using deep reinforcement learning to adjust parameters of the bidirectional long short-term memory network to form a risk prediction model and output risk prediction data;

[0050] The present invention aims to predict the behavior of defaulting on debts after a due date by using time series data constructed from a fused feature data set for risk prediction, and dynamically adjusts the time series model parameters through deep reinforcement learning, thereby constructing a risk prediction model that can output risk prediction data. Specifically, the present invention first uses multi-dimensional behavioral data to form standardized data after preprocessing, and then extracts key features from it to generate a fused feature data set; based on this fused feature data set, the system constructs continuous time series data, and uses a bidirectional long short-term memory network for training to generate risk prediction data for each user in each future time period. Subsequently, the system introduces deep reinforcement learning technology to dynamically adjust the weights of each layer in the bidirectional long short-term memory network in order to reduce the prediction error and improve the robustness of the model, and finally forms a risk prediction model optimized by reinforcement learning. The risk prediction data output by this model can be directly used for subsequent risk management and decision-making.

[0051] In practice, the method first collects various types of user behavior data, including financial transaction records, social interaction records, and device usage records. It then uses data cleaning techniques to remove abnormal data and standardizes the data to a uniform numerical range, ensuring that all data sources are compared and integrated at the same scale. Subsequently, a fixed-window feature extraction algorithm is used to perform hourly or daily statistics on the standardized data, extracting key indicators such as user account changes, social activity levels, and device usage patterns. These indicators are then accumulated according to preset weights to generate a fused feature dataset.

[0052] Next, the fused feature dataset is used as input to construct continuous time series data, which captures changes in user behavior over different time windows. To capture this temporal dynamics, a bidirectional long short-term memory (Bi-LSTM) network is used for model training. By employing forward and backward layers, Bi-LSTM can simultaneously consider the impact of both historical and future information on the current state, thereby more accurately predicting user behavioral risks. During training, the model uses a backpropagation algorithm to adjust network parameters and employs a loss function (such as mean squared error) to measure the deviation between the predicted output and the actual data, leading to continuous iterative optimization.

[0053] After initially obtaining the time series risk prediction data, the present invention further introduces deep reinforcement learning technology to dynamically adjust the parameters of the bidirectional long short-term memory network. By adopting the proximal policy optimization (PPO) algorithm, the system uses the risk prediction error as feedback input based on the preset reward function to dynamically correct the network weights and reduce the model prediction error. This process not only enables the model to adapt to the dynamic changes in user behavior, but also ensures that the model still has strong predictive ability and stability under the continuous updating of data. Ultimately, the risk prediction model formed by this optimization process can output risk prediction data as the basic data for subsequent risk management, decision-making and intervention measures. From data collection, feature fusion, time series modeling, reinforcement learning optimization to model output, this overall solution constitutes a complete data processing and prediction link for predicting overdue debt evasion behavior, providing technical support and decision-making basis for risk prevention and control.

[0054] Preferably, the time series data constructed based on the fusion feature data set is trained using a bidirectional long short-term memory network, which sets a forward layer and a backward layer in the network structure and uses a backpropagation algorithm to update the network parameters to generate time series risk prediction data for each user.

[0055] The time series data constructed based on the fused feature dataset is trained using a bidirectional long short-term memory network. Its specific implementation includes explicitly setting the forward layer and the backward layer in the network structure. The forward layer is used to process the time sequence information of the input data, while the backward layer is used to reversely capture the impact of future states on current decisions. In the specific operation, the fused feature dataset is first divided into several sub-datasets in a continuous time sequence according to a preset time window. Each sub-dataset corresponds to a fixed time interval, such as every hour or every day. The data points in each sub-dataset contain feature values ​​obtained after preprocessing and fusion of user account transaction records, social interaction records, and device usage records. These feature values ​​reflect the user's behavioral status within the time window.

[0056] After constructing the time series data, it is passed as input to a bidirectional long-short-term memory (LSTM) network. This network utilizes a standard bidirectional design. Its forward layer processes time series data by progressively passing information backward from the start time, capturing the long-term dependencies of earlier data through memory cells. The backward layer, on the other hand, passes information backward from the end of the sequence, capturing the impact of later data on the current state. The bidirectional LSTM network utilizes activation functions (such as tanh and sigmoid functions) and structures such as forget gates, input gates, and output gates to selectively memorize and forget information, automatically retaining key information that is beneficial to prediction during the information transmission process.

[0057] During the training process, the network uses the back propagation algorithm to calculate the loss function value (commonly known as mean square error, the formula is:

[0058]

[0059] Among them, y i The actual risk status, To predict the risk value, N is the number of samples. The system then determines the error between the predicted and actual outputs using gradient descent to backpropagate this error to each layer of the network to update the network parameters. During the parameter update process, the system sets a fixed learning rate and employs momentum to accelerate convergence, ensuring a stable and efficient training process.

[0060] For example, in practical applications, for a user's behavioral data over a continuous 24-hour period, the fused feature data extracted through a fixed window can generate 24 input vectors at each moment, each composed of multiple dimensions. These vectors are then fed into a bidirectional long-short-term memory network, which outputs the hidden state corresponding to each time point. These hidden states are then processed by a fully connected layer to form the final risk prediction data. This process fully leverages the temporal bidirectional dependency information captured by the bidirectional structure, ensuring that the prediction results strike a balance between considering historical trends and future dynamics.

[0061] Furthermore, during model training, the system regularly saves model weights and evaluates model performance using a validation set to ensure stable and repeatable training results. Cross-validation is used to verify the model's predictive capabilities, and training parameters are adjusted based on the validation results, further enhancing the robustness of data extraction and model training. The entire process utilizes automated data processing through programming, ensuring that model training steps are reproducible across different data collection cycles, providing solid data support and algorithmic assurance for predicting post-debt default behavior.

[0062] Preferably, Figure 2 As shown, the deep reinforcement learning adopts a proximal policy optimization algorithm to dynamically adjust the weights of the bidirectional long short-term memory network according to a preset reward function, wherein the reward function uses the risk prediction error as feedback input to form a risk prediction model and output risk prediction data.

[0063] The deep reinforcement learning adopts a proximal policy optimization algorithm to dynamically adjust the weights of the bidirectional long short-term memory network through a preset reward function. Its specific implementation involves using the risk prediction error as feedback input to achieve real-time correction of the model parameters, thereby forming a risk prediction model and outputting risk prediction data. In this technology, deep reinforcement learning is used as an adaptive adjustment mechanism. Its core principle is to continuously interact with the environment and update the policy parameters according to the reward signal to continuously reduce the prediction error. In specific implementation, the time series risk prediction data output by the bidirectional long short-term memory network is first compared with the actual risk data to calculate the risk prediction error. The error is usually measured using the mean square error or absolute error. Let the risk prediction error be Where y is the actual risk status, To predict the risk value, the preset reward function determines the reward signal based on the error value. For example, a positive reward is given when the error |e| is less than a predetermined threshold, and a negative reward is given otherwise.

[0064] In the framework of the Proximal Policy Optimization (PPO) algorithm, the policy network uses the current weights of the bidirectional long short-term memory network as the initial policy and uses the sampled risk prediction data and the corresponding reward value to calculate the policy update objective function. The updated objective function can be expressed as:

[0065]

[0066] in represents the ratio of new and old strategies, is the advantage function, and ∈ is the clipping parameter. This formula ensures that the policy update is carried out within a certain range, preventing the update amplitude from being too large and destroying the stability of the original model.

[0067] In practice, the system first generates several batches of samples through a simulated environment. Each sample contains the current state, action taken, reward received, and next state information. The advantage function is calculated based on this sample data, and the weights of the bidirectional long short-term memory network are adjusted using the Propagation-Positive-Oriented Policing (PPO) algorithm. This process uses batch gradient descent to calculate the gradient of the loss function, using automatic differentiation tools for precise calculation. The adjusted network weights are then used as the new policy in the next round of sample collection, enabling continuous iterative policy updates.

[0068] For example, the system compares predicted risk with actual risk for a specific user's behavior data over multiple consecutive time windows, calculating a series of error values. It then generates a corresponding reward signal based on a pre-set reward function and uses a proximal policy optimization algorithm to update the current policy. After multiple rounds of iteration, the model converges, significantly reducing the prediction error. This results in a highly robust risk prediction model, which produces highly credible risk prediction data for risk assessment and decision-making.

[0069] Furthermore, to ensure the stability of the deep reinforcement learning process during parameter adjustment, the system monitors the range of policy changes after each round of policy updates and sets hyperparameters to control the rate of policy change, thereby preventing the model from becoming trapped in a local optimum due to rapid convergence. The entire deep reinforcement learning optimization process is implemented in the program through modular programming and repeatedly trained and validated with actual risk data to ensure its operability and stability in predicting post-debt default behavior. Through real-time feedback and dynamic adjustment, this invention enables the bidirectional long-short-term memory network to more accurately capture subtle changes in user risk behavior, providing strong support for the output of final risk prediction data.

[0070] Based on the risk prediction data, a multi-objective decision-making algorithm is used to calculate a comprehensive risk score, and fuzzy logic is used to numerically adjust the comprehensive risk score to form decision intervention data. At the same time, an auxiliary statistical model is used to cross-validate the risk prediction data, and statistically validated risk assessment data is output;

[0071] In this invention, a comprehensive risk score is calculated based on the risk prediction data using a multi-objective decision-making algorithm. This score is then numerically adjusted using fuzzy logic to generate decision-making intervention data, ultimately providing a quantitative risk assessment result for predicting overdue debt default behavior. The core of this process lies in integrating risk assessment indicators from multiple behavioral dimensions and, through a comprehensive decision-making mechanism, forming a unified risk score. This score serves as an important basis for subsequent decision-making interventions. Specifically, the comprehensive risk score calculates the user's risk performance across various behavioral dimensions, including financial, social, and device usage, and uses a weighted cumulative approach to obtain an overall risk score. Fuzzy logic is then used to convert this score from a continuous numerical value into discrete decision parameters, which are then used to generate decision-making intervention data to guide subsequent risk control and management measures. Furthermore, the system uses auxiliary statistical models to cross-validate the risk prediction data with the decision-making intervention data to ensure the accuracy and stability of the scoring model. A five-fold cross-validation approach verifies the model's performance and ensures the validity of the risk assessment data, thus providing a solid theoretical foundation and data support for the final prediction of overdue debt default.

[0072] Preferably, the multi-objective decision-making algorithm calculates the respective risk scores for financial behavior risk, social behavior risk and device behavior risk according to preset weighting coefficients, and generates comprehensive risk score data through accumulation operation; the comprehensive risk score data is converted through a preset membership function, and the continuous risk score is discretized into decision parameters to form decision intervention data.

[0073] In the application of the multi-objective decision-making algorithm, the present invention first takes financial behavior risk, social behavior risk and device behavior risk as the main evaluation dimensions. The risk score of each dimension is calculated by a preset weighting coefficient, and the comprehensive risk score data is obtained by cumulative operation. During the calculation process, the risk score of each dimension is first normalized based on historical data to eliminate the dimensional differences between different dimensions, so that the scores of each dimension are comparable. For financial behavior risk, the system calculates by evaluating the user's credit history, debt record, overdue situation and other indicators. The specific methods include changes in credit limit, fluctuations in account balance, etc.; for social behavior risk, the system calculates the risk brought by the user's social activity by analyzing the user's interaction frequency on the social platform, content publishing frequency, social circle activity and other data; for device behavior risk, the system scores based on factors such as the user's device usage pattern, device switching frequency, and the geographical location of the login device.

[0074] During the risk score calculation process, preset weighting factors are set. For example, the weight of financial behavior risk may be 0.4, the weight of social behavior risk may be 0.3, and the weight of device behavior risk may be 0.3. The weighting factors are adjusted based on the experience of risk management experts and the performance of historical data. The relative importance of each dimension in the comprehensive score is determined through data analysis. The specific calculation process is as follows:

[0075] R total =α1×R finance +α2×R social +α3×R device

[0076] Among them, R total is the final comprehensive risk score, R finance 、R social and R device The following table represents the scores for financial, social, and device behavior risks, respectively, with α1, α2, and α3 being the corresponding weighting coefficients. Through this weighted calculation, the system can integrate the risk profiles of each dimension to generate an accurate overall risk score.

[0077] After obtaining the comprehensive risk score data, the present invention uses fuzzy logic to adjust the score and convert the continuous risk score value into a decision parameter. The role of fuzzy logic in this process is to map the original continuous risk score to a discrete decision interval through the membership function to facilitate subsequent decision intervention. The membership function is a fuzzy mathematical method that can map a numerical input to a membership value within an interval. This value represents the degree of membership of the numerical value in a specific category. For example, in the process of adjusting the risk score, assuming that the comprehensive risk score ranges from 0 to 100, the membership function can map different score values ​​to the categories of "low risk", "medium risk" and "high risk" according to preset rules.

[0078] For example, the membership function of a given object can be defined as follows:

[0079] When the risk score R total When the value is between [0,30], the membership value is the value of the "low risk" area;

[0080] When R total When the value is between [30,70], the membership value is the value of the "medium risk" area;

[0081] When R total When the value is between [70,100], the membership value is in the "high risk" area.

[0082] This adjustment process is implemented through a fuzzy rules engine, where input risk scores are classified using rules to generate decision intervention data. This decision data can be used for subsequent risk intervention measures, such as automatically generating early warning signals, triggering further review, and notifying relevant departments to take action. The core of this invention lies in the fine-tuning of risk scores through fuzzy logic, enabling the system to make flexible and appropriate decisions based on different risk levels.

[0083] Preferably, the auxiliary statistical model adopts a five-fold cross-validation method to divide the risk prediction data into a training set and a validation set, and compares the consistency between the risk prediction data and the decision intervention data by calculating statistical indicators, thereby outputting statistically verified risk assessment data.

[0084] In order to ensure the stability of the risk prediction model and the accuracy of the prediction results, the present invention uses an auxiliary statistical model for cross-validation to further verify the consistency of the risk prediction data and the decision intervention data. Cross-validation is a commonly used validation method, which evaluates the predictive ability of the model by dividing the data set into multiple subsets and using each subset as a validation set in turn, and the remaining part as a training set. In the present invention, cross-validation adopts a five-fold cross-validation method, that is, the data set is divided into 5 subsets, and one of the subsets is selected each time as the validation set, and the rest are used as training sets. Through multiple cycles of training and validation process, the stability and generalization ability of the model are finally calculated.

[0085] During cross-validation, the risk prediction model is first trained using the training set to generate a preliminary set of risk prediction data. This prediction data is then compared with the actual data to calculate the prediction error, and the model parameters are adjusted accordingly. After each round of cross-validation, the system calculates the average error across each fold to ensure consistent performance across different datasets.

[0086] Cross-validation not only verifies the accuracy of risk prediction data but also ensures a high degree of consistency in risk assessment data during the decision-making process. For example, when cross-validating financial risk scores, the system may discover significant differences in the financial behavior of certain user groups, which could affect the risk assessment results. Through cross-validation, the model can identify and correct these potential biases, ensuring that the final risk assessment data effectively reflects the user's actual behavior patterns. Ultimately, this validated data serves as the basis for decision-making interventions, providing scientific support for subsequent risk management and overdue debt processing.

[0087] Based on the statistically verified risk assessment data as feedback, the risk prediction model is updated with parameters using an incremental learning method, and the hyperparameters of the risk prediction model are adjusted using an adaptive optimization algorithm to generate a final prediction result.

[0088] In the present invention, based on statistically verified risk assessment data, the incremental learning method is used to update the parameters of the risk prediction model, and the hyperparameters of the model are adjusted through the adaptive optimization algorithm to generate the final prediction results. The combination of the incremental learning method and the adaptive optimization algorithm enables the risk prediction model to adapt to the changes in the newly collected user behavior data in real time, improving the flexibility and prediction accuracy of the model. In a specific implementation, when the new user behavior data reaches the preset threshold, the system integrates the new data into the existing model in real time through the incremental learning method, updates the model parameters, and ensures that the model can always reflect the latest user behavior characteristics. At the same time, the system uses the Bayesian optimization technology through the adaptive optimization algorithm to automatically adjust the hyperparameters of the model, further improving the generalization ability and prediction accuracy of the model in different data environments. The entire process is automatically executed without human intervention, and can continuously optimize the prediction results, and provide a highly reliable decision-making basis for the prediction of debt evasion behavior after overdue.

[0089] Preferably, the incremental learning method adopts an online gradient descent algorithm. When the newly collected user behavior data reaches a preset threshold, the new data is integrated into the risk prediction model and the model parameters are updated in real time so that the model can reflect the latest user behavior characteristics.

[0090] In implementing the incremental learning method, the present invention uses an online gradient descent algorithm to ensure that when newly collected user behavior data reaches a preset threshold, the new data can be quickly and in real time incorporated into the risk prediction model, dynamically adjusting the model parameters. The online gradient descent algorithm is a commonly used incremental learning method suitable for streaming data processing. Its core principle is to fine-tune the model parameters using only the new data as it arrives, rather than training the model from scratch each time, thereby improving computational efficiency. In the present invention, the incremental learning process is implemented through the following steps:

[0091] When new user behavior data is collected, the data is first cleaned and standardized to ensure data quality and consistency.

[0092] Use the current model to predict new data and calculate the prediction error. The mean square error (MSE) is usually used as the loss function:

[0093]

[0094] Among them, y i is the true value, is the model prediction value, and N is the number of data samples.

[0095] Use the gradient descent method to calculate the gradient of the loss function with respect to the model parameters, and adjust the parameters according to the learning rate. The formula is as follows:

[0096]

[0097] Among them, θ t is the current parameter of the model, η is the learning rate, is the gradient of the loss function with respect to the parameters.

[0098] In this way, the system makes appropriate adjustments to the existing model each time new data arrives, ensuring the timeliness and accuracy of the model while avoiding the high computational cost of training from scratch.

[0099] For example, when a user's new financial transaction data or social interaction behavior changes, the system can quickly adapt through this method and update the model parameters in real time to ensure that the model can reflect the most accurate risk prediction.

[0100] Preferably, the adaptive optimization algorithm adopts Bayesian optimization technology to automatically adjust the hyperparameters of the risk prediction model within a preset hyperparameter search range, and generates a final prediction result based on statistically verified risk assessment data as feedback.

[0101] In the application of the adaptive optimization algorithm, the present invention uses Bayesian optimization technology to automatically adjust the hyperparameters of the risk prediction model. Bayesian optimization is an efficient global optimization method that is particularly suitable for high-dimensional and complex optimization problems and can find the optimal solution with limited computing resources. The core principle of Bayesian optimization is to establish a probabilistic model of the objective function (usually using Gaussian process regression) and predict the hyperparameter configuration that is most likely to lead to the optimal result based on this model, thereby selecting the hyperparameter configuration that is most likely to improve model performance for testing in each iteration.

[0102] Specifically, in the present invention, the Bayesian optimization process can be divided into the following steps:

[0103] 1. Build a surrogate model: By sampling a subset of points in the current hyperparameter space and using the performance of these points as observations, a surrogate model is constructed. Typically, Bayesian optimization uses a Gaussian Process Regression (GPR) model to fit the objective function, representing the relationship between hyperparameters and model performance.

[0104] 2. Acquisition function: Based on the surrogate model, Bayesian optimization uses an acquisition function (such as the expected improvement (EI) function) to determine the next hyperparameter point that is most likely to improve model performance. The expected improvement function selects the next hyperparameter point by calculating the expected improvement of the current point relative to the known optimal solution:

[0105]

[0106] Where f(θ) is the objective function (such as model performance index), θ * is the currently known optimal hyperparameter configuration, and θ is the hyperparameter configuration to be tested.

[0107] 3. Evaluation and Update: Train the model for the selected hyperparameters and evaluate its performance. After obtaining new results, update the proxy model and continue to use the model to search the hyperparameter space until the stopping condition (such as the number of iterations, error threshold, etc.) is met.

[0108] Through this method, Bayesian optimization can quickly find the optimal or near-optimal configuration in a wide range of hyperparameter spaces, while avoiding redundant calculations in traditional grid search or random search methods, thereby greatly improving optimization efficiency. In the present invention, Bayesian optimization is mainly used to adjust the hyperparameters (such as learning rate, number of hidden layer units, etc.) of the long short-term memory network (LSTM) to improve the accuracy and robustness of the risk prediction model. Through adaptive adjustment, Bayesian optimization technology ensures that the model can always maintain good generalization capabilities under different data environments, and can automatically optimize model parameters and improve prediction accuracy as the data changes.

[0109] For example, if the model trained with the initial hyperparameter configuration performs poorly, Bayesian optimization can dynamically adjust parameters such as the learning rate and the number of LSTM layers, repeatedly test and optimize, and ultimately generate a model with higher prediction accuracy.

[0110] By combining incremental learning with Bayesian optimization, the present invention enables continuous optimization of risk prediction models and, based on statistically validated risk assessment data as feedback, ensures that each model update improves predictive performance. During model optimization, incremental learning continuously absorbs new data to maintain the timeliness of the model, while Bayesian optimization focuses on improving the overall performance of the model through intelligent search of the hyperparameter space. Ultimately, after multiple rounds of incremental learning and hyperparameter adjustment, the model is able to maintain stable and efficient predictive capabilities in an environment with real-time data updates, providing scientific and reliable support for the prediction of post-debt default behavior. Through the cyclic iteration of this process, the present invention effectively improves the accuracy and stability of the risk prediction model, ensuring the timeliness and accuracy of decision-making.

[0111] like Figure 3As shown, a system for predicting debt default behavior after overdue payment is used to implement the above-mentioned method for predicting debt default behavior after overdue payment, and the system includes:

[0112] The acquisition module is used to collect multi-dimensional behavioral data such as user account transaction records, user social interaction records, and user device usage records according to a preset data acquisition plan; the acquisition module uses a dedicated data acquisition board, which consists of a microcontroller and a high-speed communication interface, and integrates dedicated sensors and network interface circuits. It reads multi-dimensional behavioral data in real time from the user account transaction system, social platform server, and device log interface through a preset protocol, and uses a buffer memory to temporarily cache the data to ensure that the data is not lost during transmission. The acquisition module passes the complete data packet to the next processing unit.

[0113] The data processing module is used to clean the multi-dimensional behavioral data, remove abnormal data, fill in missing data, and perform minimum and maximum normalization processing. It also extracts features from the normalized data to generate a fused feature dataset. The data processing module is based on a hardware platform composed of a field programmable gate array and a digital signal processor, which implements data cleaning, abnormal data removal, missing data filling, and minimum and maximum normalization. This module embeds a dedicated algorithm core, which uses hardware logic to perform high-speed filtering and statistical analysis on the input data. It also uses built-in memory to store normalization parameters and perform unified data conversion. On this basis, the system also uses a hardware accelerator to calculate the features extracted from the raw data, achieving rapid feature fusion and generating a high-quality fused feature dataset. The processing results are then transmitted to the time series modeling module via a high-speed interface.

[0114] The time series modeling module is used to construct time series data based on the fused feature data set, and use a bidirectional long short-term memory network to train the time series data to generate time series risk prediction data for each user. At the same time, deep reinforcement learning is used to adjust the parameters of the weights of the bidirectional long short-term memory network to form a risk prediction model and output risk prediction data. The time series modeling module mainly relies on a graphics processing unit or a dedicated deep learning accelerator card to realize the hardware implementation of the bidirectional long short-term memory network (Bi-LSTM). This module organizes the fused feature data set output by the data processing module into continuous time series data, stores and schedules the data in chronological order through a dedicated memory, and uses a pre-programmed neural network accelerator to perform forward propagation and backpropagation calculations to generate time series risk prediction data for each user in real time. At the same time, the embedded deep reinforcement learning engine uses a proximal policy optimization algorithm to adjust the network parameters in real time to reduce the prediction error, form a hardware-optimized risk prediction model, and pass the prediction data to the decision module through a high-speed data bus.

[0115] The decision-making module is used to calculate the risk score of each risk dimension based on the risk prediction data using a multi-objective decision-making algorithm, calculate the comprehensive risk score according to the preset weighting coefficient, and perform numerical conversion of the comprehensive risk score through a preset membership function to form decision intervention data. At the same time, the risk prediction data is cross-validated using an auxiliary statistical module to output statistically verified risk assessment data. The decision-making module is integrated into a system-level chip and is composed of an embedded processor and a custom logic circuit to implement a multi-objective decision-making algorithm. The module performs weighted calculations on the risk prediction data output by the time series modeling module according to the financial, social, and device risk dimensions, and generates comprehensive risk score data through preset weighting coefficients and cumulative operations. Subsequently, the preset membership function implemented in hardware converts the continuous score into discrete decision parameters to form decision intervention data. In addition, the decision-making module has a built-in auxiliary statistical unit that uses a cross-validation algorithm to group and compare the risk prediction data and calculate statistical indicators to ensure that the output risk assessment data has been hardware-verified and meets the system decision requirements.

[0116] The model update module is used to update the parameters of the risk prediction model using an online gradient descent incremental learning method based on the risk assessment data feedback, and automatically adjust the hyperparameters of the risk prediction model within a preset search interval using Bayesian optimization technology to generate the final prediction result. The model update module uses the online gradient descent accelerator and Bayesian optimization engine integrated in the system to update the risk prediction model in real time. The module receives the risk assessment feedback data returned by the decision module through a high-speed interface. When the newly collected data reaches the preset quantization threshold, the updater immediately triggers the online gradient descent algorithm to fine-tune the current model parameters; at the same time, the Bayesian optimization technology is used to automatically search for the optimal configuration within the preset hyperparameter search interval, adjust the model hyperparameters, and achieve adaptive optimization. The updated model parameters are fed back to the time series modeling module in real time through memory sharing and high-speed communication, and the optimized final prediction result is finally generated, ensuring that the system always maintains prediction accuracy and response speed in actual applications.

[0117] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware.

[0118] The above are merely embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included within the scope of the claims of the present application.

Claims

1. A method for predicting debt default behavior after overdue payment, characterized in that: The following steps are involved: Collect multi-dimensional behavioral data of users, perform data cleaning and data standardization on the behavioral data, extract features from the standardized data, and generate a fused feature data set; The multi-dimensional behavioral data is collected from user account transaction records, user social interaction records, and user device usage records in a preset data format. An outlier detection algorithm is used to remove data that exceeds the statistical range. A missing data filling algorithm is used to compensate for missing items in the data. The collected data is normalized using a minimum and maximum normalization method to generate standardized raw data with a unified numerical range. The features in the standardized raw data are calculated using a fixed-window-based feature extraction algorithm. The calculation process quantifies the user's account changes, social interaction frequency, and device usage patterns, and the quantified results are accumulated according to a preset formula to form a fused feature dataset; Constructing time series data based on the fused feature dataset, using a bidirectional long short-term memory network for training to generate time series risk prediction data for each user, and using deep reinforcement learning to adjust parameters of the bidirectional long short-term memory network to form a risk prediction model and output risk prediction data; The deep reinforcement learning adopts a proximal policy optimization algorithm to dynamically adjust the weights of the bidirectional long short-term memory network according to a preset reward function, where the reward function uses the risk prediction error as feedback input to form a risk prediction model and output risk prediction data; Based on the risk prediction data, a multi-objective decision-making algorithm is used to calculate a comprehensive risk score, and fuzzy logic is used to numerically adjust the comprehensive risk score to form decision intervention data. At the same time, an auxiliary statistical model is used to cross-validate the risk prediction data, and statistically validated risk assessment data is output; Based on the statistically verified risk assessment data as feedback, the risk prediction model is updated with parameters using an incremental learning method, and the hyperparameters of the risk prediction model are adjusted using an adaptive optimization algorithm to generate a final prediction result.

2. The method according to claim 1, characterized in that The time series data constructed based on the fusion feature data set is trained using a bidirectional long short-term memory network. The bidirectional long short-term memory network sets a forward layer and a backward layer in the network structure, and uses a backpropagation algorithm to update the network parameters to generate time series risk prediction data for each user.

3. The method according to claim 1, characterized in that The multi-objective decision-making algorithm calculates the risk scores of financial behavior risk, social behavior risk, and device behavior risk according to preset weighting coefficients, and generates comprehensive risk score data through cumulative calculation; The comprehensive risk score data is converted through a preset membership function to discretize the continuous risk score into decision parameters to form decision intervention data.

4. The method according to claim 1, wherein The auxiliary statistical model adopts a five-fold cross-validation method to divide the risk prediction data into a training set and a validation set, and compares the consistency between the risk prediction data and the decision intervention data by calculating statistical indicators, thereby outputting statistically verified risk assessment data.

5. The method according to claim 1, characterized in that The incremental learning method adopts an online gradient descent algorithm. When the newly collected user behavior data reaches a preset threshold, the new data is integrated into the risk prediction model and the model parameters are updated in real time.

6. The method according to claim 1, characterized in that The adaptive optimization algorithm uses Bayesian optimization technology to automatically adjust the hyperparameters of the risk prediction model within a preset hyperparameter search range, and generates a final prediction result based on statistically verified risk assessment data as feedback.

7. A system for predicting overdue debt evasion behavior, used to implement the method for predicting overdue debt evasion behavior according to any one of claims 1 to 6, characterized in that: The system includes: The collection module is used to collect multi-dimensional behavioral data such as user account transaction records, user social interaction records, and user device usage records according to a preset data collection plan; A data processing module is used to perform data cleaning, abnormal data removal, missing data filling, and minimum and maximum normalization on the multi-dimensional behavior data, and extract features from the normalized data to generate a fused feature data set; A time series modeling module is used to construct time series data based on the fused feature dataset, train the time series data using a bidirectional long short-term memory network, generate time series risk prediction data for each user, and adjust the weight parameters of the bidirectional long short-term memory network using deep reinforcement learning to form a risk prediction model and output risk prediction data; A decision-making module is configured to calculate risk scores for each risk dimension using a multi-objective decision-making algorithm based on the risk prediction data, calculate a comprehensive risk score according to a preset weighting coefficient, perform numerical conversion on the comprehensive risk score using a preset membership function to form decision intervention data, and simultaneously use an auxiliary statistical module to cross-validate the risk prediction data and output statistically validated risk assessment data; The model updating module is used to update the parameters of the risk prediction model based on the risk assessment data feedback using the online gradient descent incremental learning method, and to automatically adjust the hyperparameters of the risk prediction model within a preset search interval using the Bayesian optimization technology to generate the final prediction result.

Images