Fraud detection method based on SMOTE-CatBoost-SHAP imbalanced dataset

Abstract

The present application relates to the technical field of risk management and data analysis, in particular to a fraud detection method and system for imbalanced data sets, more specifically to a fraud detection method based on SMOTE-CatBoost-SHAP imbalanced data sets; comprising the following steps: (1) data preprocessing step; (2) model training step; (3) model evaluation step; (4) model interpretation step; the present application comprehensively applies data balancing technology, advanced machine learning algorithms and explainability framework, forming a complete fraud detection solution; this comprehensive technical advantage enables the model to simultaneously achieve high-precision fraud identification and explainable decision support when processing complex and imbalanced trade data, providing a new technical path and solution for the field of risk fraud detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of risk management and data analysis, in particular to a fraud detection method and system for imbalanced data sets, and more specifically to a fraud detection method based on SMOTE-CatBoost-SHAP imbalanced data sets. BACKGROUND

[0002] In the field of risk fraud detection, due to the concealment and complexity of fraudulent behavior, and the fact that fraudulent samples usually only account for a small proportion of the overall data set, this leads to a serious problem of data imbalance. This data imbalance poses a challenge to the training of machine learning models, as traditional classification algorithms are prone to overfitting or underfitting when faced with imbalanced data sets, thereby affecting the accuracy and generalization ability of the model. In particular, in the application scenario of import and export trade fraud detection, with the increase in global trade volume and the diversification of fraudulent means, how to effectively identify and prevent fraudulent behavior has become a key problem that customs and regulatory agencies need to solve urgently.

[0003] Traditional fraud detection methods, such as rule-based expert systems and statistical methods, can identify fraudulent behavior to some extent, but have many limitations. Expert systems rely on manually set rules and are difficult to cope with the diversity and dynamic changes of fraudulent behavior. Statistical methods often assume that data follows a specific distribution, which is not always true in complex trade data. Therefore, the effectiveness of these methods in practical applications is limited.

[0004] With the rapid development of machine learning technology, fraud detection methods based on machine learning have gradually emerged. Machine learning models can learn and identify complex fraud patterns from historical transaction data through big data analysis, improving the efficiency and accuracy of fraud detection. However, on imbalanced data sets, machine learning models also face challenges. To address this issue, scholars have made numerous attempts from the data level, algorithm level, and combination of data and algorithm level, such as using undersampling, oversampling, and comprehensive sampling methods to balance the data set, or using cost-sensitive learning and Boosting algorithms to optimize model performance.

[0005] Among them, SMOTE (Synthetic Minority Over-sampling Technique) as an effective oversampling method, balances the data set by synthesizing minority class samples, and is widely used in classification problems of imbalanced data sets. CatBoost, as a model based on gradient boosting decision trees, optimizes class feature processing and avoids overfitting, and exhibits superior performance in classification problems.

[0006] However, machine learning-based fraud detection models still face the problem of poor explainability; these models are often considered black boxes, making it difficult to explain their decision-making process, leading to regulatory agencies' lack of trust in and low acceptance of fraud identification conclusions output by the models; to improve the explainability of the model, the SHAP (SHapley Additive exPlanations) method has emerged.

[0007] SHAP is a model explanation method based on Shapley values in game theory, which can assign a contribution value to each feature representing the impact of the feature on the model prediction; by calculating the SHAP value of each sample, it can intuitively understand how the model makes decisions, thereby improving the explainability and transparency of the model; combining SHAP method with machine learning model can reveal the key features affecting fraud transaction identification, providing more reliable fraud detection basis for regulatory agencies. SUMMARY

[0008] The present application aims to provide a fraud detection scheme based on SMOTE-CatBoost-SHAP, which aims to solve the problem of model performance decline caused by severe data imbalance in import and export trade fraud detection, and enhance the explainability of the model by introducing the SHAP (SHapley Additive exPlanations) method, improve the accuracy and credibility of fraud detection; this technical solution combines data balancing technology (SMOTE), advanced machine learning algorithm (CatBoost) and model explanation tool (SHAP), designed specifically for handling complex and highly imbalanced trade fraud data, which is of great significance for promoting international trade safety, improving customs supervision efficiency and promoting the application of machine learning technology in risk management.

[0009] The technical solution adopted by the present application to solve the technical problem is: a fraud detection method based on SMOTE-CatBoost-SHAP non-uniform data set, comprising the following steps:

[0010] (1) Data preprocessing step: obtain the original data set and perform standardization processing, then divide the data set into training data set and test data set;

[0011] (2) Model training step: use the training data set, combine with SMOTE algorithm for data balancing, then train CatBoost fraud detection model;

[0012] (3) Model evaluation step: use the test data set to verify the identification performance of the fraud detection model based on SMOTE-CatBoost;

[0013] (4) Model explanation step: introduce SHAP method, calculate the contribution value of each feature to the model prediction result, identify the key features that have important contribution to the fraud detection identification result.

[0014] The data preprocessing step also includes outlier detection and missing value processing operations on the original data set.

[0015] The SMOTE algorithm balances the data set by randomly selecting minority class samples and synthesizing new minority class samples in data balancing processing.

[0016] The CatBoost fraud detection model uses the ranking boosting strategy in the training process to ensure that only part of the data before the current data is used to calculate the residual when building each tree, avoiding the problem of data leakage.

[0017] The CatBoost fraud detection model uses target smoothing encoding technology for category features, which encodes category features as statistical information with target variables, avoiding the problem of dimension expansion caused by one-hot encoding.

[0018] In the model evaluation step, accuracy, recall rate, F1 score and other indicators are used to evaluate the identification performance of the SMOTE-CatBoost-based fraud detection model.

[0019] The SHAP method in the model explanation step calculates the Shapley value of each feature to quantify the marginal contribution of each feature to the prediction result and provide transparency of the model decision-making process.

[0020] The SHAP method also includes visualizing Shapley values to visually display which variables have higher weights when judging fraud risk.

[0021] A fraud detection method based on SMOTE-CatBoost-SHAP non-uniform data set also includes a comparison step: comparing the performance of the SMOTE-CatBoost-based fraud detection model with other machine learning models such as logistic regression, decision tree classifier, gradient boosting classifier, K-nearest neighbor classifier, naive Bayes classifier, multilayer perceptron, stochastic gradient descent classifier, etc. to verify the superiority of the proposed model.

[0022] The fraud detection method based on SMOTE-CatBoost-SHAP non-uniform data set can be applied to import and export trade fraud detection scenarios, and verified using real customs declaration commodity data sets to prove its effectiveness and practicality.

[0023] The beneficial effects of the technical solution are: the application comprehensively applies data balancing technology, advanced machine learning algorithms and explainability framework, and forms a complete fraud detection solution; the comprehensive technical advantages enable the model to achieve high-precision fraud identification and explainable decision support when processing complex and unbalanced trade data, providing a new technical path and solution for risk fraud detection, which is embodied as follows:

[0024] In the face of changing fraud methods, the model proposed by the application shows good adaptability and flexibility; through SHAP analysis, new fraud patterns or new varieties of known fraud methods can be discovered and explained in a timely manner, providing strong decision support for customs and regulatory agencies; this ability to dynamically adjust and optimize fraud detection strategies is unmatched by traditional fraud detection systems based on rules or statistical methods.

[0025] The embodiments of the application take import and export trade fraud detection as an application scenario, and verify the classification accuracy and explainability of the model through real customs declaration commodity data sets; the experimental results show that the model can not only effectively identify fraudulent behavior, but also provide clear explanation and visual analysis, greatly improving the value of the model in practical application; this has important significance for improving the efficiency of customs supervision, cracking down on trade fraud, and maintaining fair trade order. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is an execution logic principle schematic diagram in the embodiment.

[0027] Figure 2 is a feature selection result schematic diagram based on a decision tree in the embodiment.

[0028] Figure 3 is a SHAP feature analysis schematic diagram in the embodiment.

[0029] Figure 4 is a waterfall chart of SHAP values in the embodiment. DETAILED DESCRIPTION

[0030] The specific implementation of the technical solution will be described in detail below in combination with specific embodiments:

[0031] Embodiment 1: applied to import and export declaration enterprise fraud detection scene

[0032] With the continuous growth of global trade, the number and variety of imported and exported goods are increasing, and import and export declaration has become an important link in international trade. However, trade fraud in import and export declaration has also become prominent, becoming a key risk area for governments and regulatory agencies. According to the report of the World Customs Organization (WCO), trade fraud not only seriously affects customs revenue, but also can lead to unfair competition and market confusion. In recent years, with the rapid growth of cross-border trade complexity and transaction volume, fraud methods have become increasingly diverse, including false declaration of commodity value, false commodity classification, and false origin. Import and export declaration fraud refers to the behavior of intentionally false declaration of commodity information by enterprises or individuals to evade customs or obtain improper preferential treatment for illegal economic benefits. These fraudulent behaviors have become more difficult to detect with the increase of global trade activities. Traditional fraud detection methods face many challenges, such as large transaction data size, complex data dimensions, and dynamic changes in fraud patterns. Therefore, how to effectively identify fraudulent behaviors in import and export declaration using advanced technology has become a key problem for customs and regulatory agencies to solve.

[0033] This embodiment will be based on SMOTE-CatBoost-SHAP to detect fraud in import and export declaration of enterprises facing serious non-equilibrium data sets. As shown in Figure 1 The execution practice of this embodiment is divided into the following four parts:

[0034] (1) Data collection and preprocessing

[0035] First, the original data is obtained, and the data set is standardized. Then, the constructed data set is proportionally divided into training data set and test data set.

[0036] (2) Model training and optimization

[0037] This embodiment trains the SMOTE-CatBoost model based on the training data set, and compares SMOTE-LR, SMOTE-DTC, SMOTE-GBC, SMOTE-KNNC, SMOTE-BNB, SMOTE-GNB, SMOTE-MLPC and SMOTE-SGDC.

[0038] (3) Model testing and performance evaluation

[0039] This embodiment uses the test data set to verify the identification performance of the import and export declaration enterprise fraud detection model based on SMOTE-CatBoost.

[0040] (4) Model explainable analysis

[0041] This embodiment uses the SHAP method to identify the key features that have important contributions to the identification results of import and export declaration enterprise fraud detection.

[0042] The main purpose of this embodiment in the implementation of the experiment is to build an import and export declaration enterprise fraud detection model based on CatBoost; CatBoost is a model based on Gradient Boosting Decision Trees (GBDT), which is specially optimized for category feature processing and overfitting avoidance; The main innovation of CatBoost lies in Ordered Boosting and native category feature processing; Ordered Boosting avoids the common data leakage problem in Gradient Boosting by dividing the data into sequential blocks, ensuring that only the part before the current data is used to calculate the residual when building each tree; Especially when dealing with time series data or small sample datasets, ordered boosting can significantly improve the generalization performance of the model.

[0043] For category features, CatBoost does not require traditional one-hot encoding or target encoding, but uses a special target smoothing encoding technique; It is based on the correlation between category features and target variables, by encoding category features into their statistical information with target variables, and combining ordered boosting to prevent information leakage, which not only maintains the information integrity of category features, but also avoids the dimension expansion problem caused by one-hot encoding; The function form of CatBoost is:

[0044]

[0045] Where Tk(x) represents the prediction of the kth decision tree, ak is the weight of the kth tree, and K is the total number of trees in the model; The final prediction value f(x) is obtained by the weighted sum of all decision trees.

[0046] Then fuse the model interpretability of SHAP; SHAP is an algorithm for explaining machine learning models, which assigns a contribution value to each feature based on the Shapley value in game theory, indicating the influence of the feature on the model prediction; SHAP provides a consistent and unified method to explain the predictions of any machine learning model, and can quantify the marginal contribution of each feature to the prediction result; The theoretical basis of SHAP algorithm comes from the Shapley value in game theory; Shapley value is a fair method to allocate the benefits of each participant in cooperative games, in machine learning, features can be regarded as participants in games, and model predictions can be regarded as benefits; Shapley value determines the importance of each feature in the current sample by calculating the marginal contribution of the feature in all possible feature combinations; Specifically, for each sample x, its model prediction f(x) is decomposed into the sum of the contributions of each feature:

[0047]

[0048] Where, is the baseline value (usually the average prediction value of the model on the scene data); is the Shapley value of feature i, which indicates the marginal contribution of feature i to the prediction result.

[0049] The calculation of Shapley value involves traversing all possible subsets of features and calculating marginal contributions; for a feature i, the Shapley value The formula is:

[0050]

[0051] Where N is the set of all features, S is a subset of feature i, excluding i itself, f(s) represents the predicted value of the model when only the feature subset S is used; f(s∪{i}) represents the predicted value after adding feature i to the feature subset S. is the weight assigned to each subset in the Shapley value, ensuring that the marginal contribution of each feature can be calculated fairly.

[0052] This embodiment combines SMOTE data balancing technology with the CatBoost machine learning algorithm to effectively address the overfitting or underfitting issues of fraud detection models in unbalanced datasets. SMOTE technology balances the dataset by synthesizing minority class samples, while the CatBoost algorithm further enhances the model's fraud detection capabilities in complex trade data through its unique ranking boosting and native category feature processing techniques. Compared with existing models such as logistic regression, decision tree classifiers, and gradient boosting classifiers, the SMOTE-CatBoost model proposed in this embodiment demonstrates higher classification accuracy and generalization performance in fraud detection tasks. Furthermore, this invention innovatively introduces the SHAP interpretation framework, providing powerful interpretability support for the SMOTE-CatBoost model. SHAP calculates the contribution of each feature to the model's prediction results (SHAP value), enabling researchers and business personnel to intuitively understand which variables carry higher weight when determining fraud risk. This transparent interpretation not only enhances the model's credibility but also helps business experts more accurately identify key influencing factors of fraudulent behavior, thereby optimizing fraud detection strategies.

[0053] Example 2: Import and export trade fraud detection model based on SMOTE-CatBoost-SHAP and interpretability improvement method

[0054] The embodiment aims to combine data balancing techniques with machine learning to weaken the above-mentioned phenomenon and improve model performance. In addition, when performing risk assessment and disposal on model detection results, business experts pay more attention to the explainability of black-box models. The embodiment introduces SHAP (SHapley Additive exPlanations) to identify important influencing factors of fraudulent behavior based on the above-mentioned model architecture, thereby improving the explainability and scenario usability of the model. The research takes import and export trade fraud detection as an application scenario, and uses real customs declaration data set to verify the classification accuracy and explainability of the risk fraud detection model SMOTE-CatBoost-SHAP proposed in the embodiment.

[0055] The embodiment is based on SMOTE-CatBoost-SHAP for import and export declaration enterprise fraud detection of serious non-balanced data set. The research framework is as shown in Figure 1 The embodiment is based on SMOTE-CatBoost-SHAP for import and export declaration enterprise fraud detection of serious non-balanced data set. The research framework is as shown in

[0056] (1) Data collection and preprocessing; first, the original data is obtained, and the data set is standardized; then, the constructed data set is proportionally divided into training data set and test data set.

[0057] (2) Model training and optimization; the embodiment trains the SMOTE-CatBoost model based on the training data set, and compares SMOTE-LR, SMOTE-DTC, SMOTE-GBC, SMOTE-KNNC, SMOTE-BNB, SMOTE-GNB, SMOTE-MLPC and SMOTE-SGDC.

[0058] (3) Model testing and performance evaluation; the test data set is used to verify the recognition effect of the import and export declaration enterprise fraud detection model based on SMOTE-CatBoost.

[0059] (4) Model explainable analysis; the embodiment uses SHAP method to identify the key features that have important contribution to the import and export declaration enterprise fraud detection recognition result.

[0060] In this embodiment, CatBoost is a model based on Gradient Boosting Decision Trees (GBDT), which is specially optimized for handling categorical features and avoiding overfitting; the main innovation of CatBoost lies in Ordered Boosting and native handling of categorical features; Ordered Boosting avoids the common data leakage problem in Gradient Boosting by dividing the data into sequential blocks, ensuring that only the part before the current data is used to calculate the residual when building each tree, thus significantly improving the generalization performance of the model, especially when dealing with time series data or small sample datasets.

[0061] For categorical features, CatBoost does not require traditional one-hot encoding or target encoding, but uses a special target smoothing encoding technique; it is based on the correlation between categorical features and target variables, by encoding categorical features into their statistical information with target variables, and combining Ordered Boosting to prevent information leakage, which not only maintains the information integrity of categorical features, but also avoids the dimension expansion problem caused by one-hot encoding; the function form of CatBoost is:

[0062]

[0063] where T k (x) represents the prediction of the kth decision tree, α k is the weight of the kth tree, and K is the total number of trees in the model; the final prediction value f(x) is obtained by weighted sum of all decision trees.

[0064] SHAP is an algorithm for explaining machine learning models, which assigns a contribution value to each feature based on the Shapley value in game theory, indicating the influence of the feature on the model prediction, and can quantify the marginal contribution of each feature to the prediction result.

[0065] The theoretical basis of SHAP algorithm comes from the Shapley value in game theory; the Shapley value is a fair method for distributing the benefits of each participant in a cooperative game, in machine learning, features can be regarded as participants in the game, and the model's prediction value can be regarded as benefits; the Shapley value determines the importance of each feature in the current sample by calculating the marginal contribution of the feature in all possible feature combinations; specifically, for each sample x, its model prediction f(x) is decomposed into the sum of the contributions of each feature:

[0066]

[0067] where, is the baseline value (usually the average prediction value of the model on the background data); is the Shapley value of feature i, representing the marginal contribution of feature i to the prediction result.

[0068] The calculation of the Shapley value involves traversing all possible subsets of features and calculating marginal contributions; the Shapley value of a feature i The formula is:

[0069]

[0070] where N is the set of all features, S is a subset of features i, excluding i itself,

[0071] f(S) represents the prediction value of the model using only the feature subset S; f(S∪{i}) represents the prediction value after adding feature i to the feature subset S, is the weight assigned to each subset in the Shapley value, ensuring that the marginal contribution of each feature can be calculated fairly.

[0072] Based on this, the following will be further demonstrated and illustrated through specific experiments:

[0073] Through the mining and exploration of the big data resources mastered by the customs, 1 million registered import and export trade enterprises are collected, with four dimensions of enterprise basic information, transaction information, operating status, and violation records as the basic direction, and basic information including enterprise, personnel, transaction, and past penalty records are selected; high-risk, medium-risk, and low-risk samples are 33406, 46327, and 920267 respectively.

[0074] In order to ensure the performance and stability of the model, the original data is systematically preprocessed in this embodiment; first, the missing values are filled, the abnormal values are detected and processed, the samples with mixed data types are removed, and the classification variables are processed by one-hot encoding; in order to facilitate model identification and use of time information, time stamp conversion processing is performed on time series data; secondly, in order to solve the class imbalance problem in the data set, the SMOTE(Synthetic Minority Over-sampling Technique) method is used to process the data, SMOTE generates new synthetic samples between minority class samples, effectively increasing the number of minority class samples, thereby balancing the class distribution; this embodiment selects 500,000 low-risk enterprises, as well as 46327 medium-risk enterprises and 33406 high-risk enterprises, and balances the data sets of high, medium, and low risk enterprises through SMOTE technology, obtaining 500,000 enterprise data, and then dividing the training set and test set according to the 7:3 ratio.

[0075] To improve model performance, this embodiment uses feature selection engineering to eliminate irrelevant or redundant features, thereby reducing the number of features, improving the accuracy of the fraud detection model for import and export declaration companies, and reducing running time. The tree model can select features to take into account the interaction between features, process nonlinear features and high-dimensional data, and is suitable for data sets with complex relationships between features. Therefore, this embodiment selects a decision tree as the evaluation model for feature selection.

[0076] The goal of feature selection is to maximize the accuracy of the model calculated in the five-fold cross validation by optimizing the feature combination. The model uses an iterative cycle method to gradually adjust the selected feature ratio to find the best feature subset; the final result is as follows Figure 2 As shown, the horizontal axis represents the percentage of the selected features in the total features, and the vertical axis represents the accuracy of the model under this feature ratio; the study found that as the number of selected features increases, the accuracy of the model also increases; in order to make full use of the information in the data and improve the prediction performance of the model, this embodiment screened a total of 25 features through feature engineering to ensure that the model can capture the important patterns and laws in the data to the greatest extent.

[0077] In order to objectively evaluate the effectiveness of the algorithm, three evaluation indicators, namely accuracy, recall rate, and F1 value, are used to measure the model performance; the evaluation indicators are all based on the confusion matrix, as shown in Table 1 below:

[0078]

[0079]

[0080] Table 1 Confusion matrix

[0081]

[0082] Among them, True Positive (TP) represents the number of correctly identified samples; False Positive (FP) represents the number of incorrectly identified samples; True Negative (TN) represents the number of correctly identified samples as incorrect; False Negative (FN) represents the number of incorrectly identified samples as correct.

[0083] To verify the effectiveness of the SMOTE-CatBoost-based fraud identification model proposed in this embodiment, a comparative analysis was performed with the SMOTE-LR, SMOTE-DTC, SMOTE-GBC, SMOTE-KNNC, SMOTE-BNB, SMOTE-GNB, SMOTE-MLPC, and SMOTE-SGDC models. The comparison results of various models in terms of precision, recall, and F1 value are shown in Table 2.

[0084] From the accuracy of fraud identification of each model after balancing the data set, the fraud identification accuracy of SMOTE-CatBoost is the highest, with an accuracy of 84.8%, and the fraud identification accuracy of SMOTE-BNB is the lowest, with an accuracy of 8.1%; Specifically, the identification accuracy of the SMOTE-CatBoost model is 66.9%, 3.8%, 4.0%, 15.4%, 7.5%, 946.9%, 115.2% and 108.4% higher than that of the SMOTE-LR, SMOTE-DTC, SMOTE-GBC, SMOTE-KNNC, SMOTE-BNB, SMOTE-GNB, SMOTE-MLPC and SMOTE-SGDC eight fraud identification models, respectively.

[0085]

[0086]

[0087] Table 2 Evaluation index of different models

[0088] The fraud detection and identification of import and export declaration enterprises based on SMOTE-CatBoost has higher prediction accuracy, but since it is a "black box" model, it is difficult for the risk management department to understand which variables play a key role; To solve this problem, the SHAP framework is introduced to explain the model results, which provides the importance ranking of features and focuses on explaining the specific impact of different feature values on the prediction results; As shown in Figure 3 , each point in the figure represents a sample, and the sample point changes from blue to red, indicating that the SHAP value of the sample changes from small to large; The image of each index is composed of the SHAP values of all sample points, and the SHAP value equal to zero is the middle dividing line; The sample points on the left side have negative SHAP values for the index, i.e. the index has a negative contribution to the fraud identification of import and export declaration enterprises; The sample points on the right side have positive SHAP values for the index, i.e. the index has a positive contribution to the fraud detection and identification of import and export declaration enterprises; It can be seen that the features based on SHAP values can accurately analyze the influence of each sample and each feature on the fraud detection and identification of import and export declaration enterprises.

[0089] As shown in Figure 3It can be seen that the SHAP value fluctuation of the effective date, inspection rate, and annual review date is large, indicating that these features have a significant impact on the model's judgment. Among them, the annual review date and the effective date have a negative impact on the evaluation of enterprise fraud risk, that is, the later the annual review date and the effective date, the lower the fraud risk of the enterprise, which is consistent with the logic in reality. The later the annual review date and the effective date, the lower the fraud risk of the enterprise, which is consistent with the logic in reality. The higher the inspection rate, the lower the fraud risk of the enterprise, which may be because increasing supervision and constraints reduces the risk expectations of enterprises, prompting them to comply with rules and regulations and reducing the probability of fraudulent behavior.

[0090] In this embodiment, the waterfall chart is an effective visualization tool designed to visually display the contribution of each feature to the model's prediction results. It comprehensively presents the predicted value of a specific sample, the influencing factors, the average predicted value of all features, and the average marginal contribution of each feature. In this chart, the positive or negative impact of each feature on the final prediction result is displayed for a specific sample's model output. According to the analysis results of the sample waterfall chart, it can be seen that "total tax amount" and "enterprise level" are the features that have the greatest positive impact on the prediction results, significantly improving the model's prediction value. In contrast, "enterprise credit sign" shows the greatest negative impact, suppressing the improvement of the prediction results. The forces of all features eventually converge to form the model's prediction results, fully reflecting the interaction and importance of features in the decision-making process.

[0091] This embodiment uses public data to model and analyze import and export declaration enterprise fraud detection, and the following conclusions are clearly obtained:

[0092] (1) The embodiment constructs an import and export declaration enterprise fraud detection model based on SMOTE-CatBoost. To verify the performance of the model, a comparative experiment is conducted with seven classification models, and the results show that the recognition accuracy of the import and export declaration enterprise fraud detection based on SMOTE-CatBoost is the highest, and the recognition accuracy of the import and export declaration enterprise fraud detection based on SMOTE-BNB is the lowest.

[0093] (2) To verify the importance of different variables on the results. Through SHAP method, the feature importance analysis of the overall model shows that the effective date, inspection rate, and annual review date have a greater impact. The SHAP model interpretation result shows that the annual review date, effective date, and inspection rate have a negative significant impact, providing a reference for the key monitoring indicators of import and export declaration enterprise fraud detection.

Claims

1. A fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset, characterized by: The following steps are involved: (1) Data preprocessing step: obtain the original data set, perform standardization, detect outliers and process missing values ​​on the original data set, and then divide the data set into a training data set and a test data set; (2) Model training steps: Using the training dataset, combined with the SMOTE algorithm to perform data balancing, the dataset is balanced by randomly selecting minority class samples and synthesizing new minority class samples, and then the CatBoost fraud detection model is trained. During the training process, the CatBoost fraud detection model uses a ranking boosting strategy to ensure that only the part before the current data is used to calculate the residual when building each tree, thereby avoiding data leakage. The target smoothing coding technology is used for categorical features, encoding the categorical features as statistical information about them and the target variable, thereby avoiding the dimensionality expansion problem caused by one-hot encoding. (3) Model testing and performance evaluation steps: Use the test dataset to verify the recognition performance of the import and export declaration enterprise fraud detection model based on SMOTE-CatBoost; (4) Model interpretation step: The SHAP method is introduced to calculate the contribution value of each feature to the model prediction result. The Shapley value determines the importance of each feature in the current sample by calculating the marginal contribution of the feature in all possible feature combinations; based on this, the key features that have an important contribution to the fraud detection recognition results are identified.

2. The fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset according to claim 1, characterized in that: In the model evaluation step, three indicators, namely precision, recall and F1 score, are used to evaluate the recognition performance of the fraud detection model based on SMOTE-CatBoost.

3. The fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset according to claim 1, characterized in that: In the model interpretation step, the SHAP method calculates the Shapley value of each feature to quantify the marginal contribution of each feature to the prediction result, thereby providing transparency in the model decision-making process.

4. The fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset according to claim 3, characterized in that: The SHAP method also includes visualizing Shapley values ​​to intuitively show which variables have higher weights in judging fraud risk.

5. The fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset according to claim 1, characterized in that: It also includes a comparison step: comparing the performance of the SMOTE-CatBoost-based fraud detection model with other machine learning models to verify the superiority of the proposed model.

6. A fraud detection method based on SMOTE-CatBoost-SHAP unbalanced dataset according to any one of claims 1 to 5, characterized in that: This method is applied to the import and export trade fraud detection scenario and verified using a real customs declaration commodity dataset to prove its effectiveness and practicality.

Images