基于集成学习的金融交易异常检测方法OA

With the rapid development of financial technology,the volume of transaction data has grown significantly,while abnormal behaviors such as fraud,money laundering,and illegal fundraising have become increasingly frequent,posing serious threats to the security and stability of financial systems.To address these challenges,we propose an ensemble learning approach that integrates multiple machine learning models to enhance anomaly detection in financial transactions.Specifically,the proposed method combines three gradient boosting decision tree models—XGBoost,LightGBM,and CatBoost—using soft voting and stacking strategies to improve predictive accuracy and model robustness.To mitigate the impact of class imbalance during training,the SMOTE oversampling technique is employed to expand minority class samples.In addition,systematic data preprocessing,including data cleaning,feature engineering,and normalization,is applied to improve the model's generalization ability across diverse financial scenarios.Experiments conducted on five public financial datasets,including the CCF illegal fundraising dataset,demonstrate that the ensemble strategy,especially the soft voting method,outperforms individual models in terms of F1 score,AUC,and stability.The study also identifies promising directions for future work,such as improving model interpretability,enabling real-time detection,and exploring the integration of federated learning and graph neural networks.Overall,the proposed method shows strong practical adaptability and holds significant potential for application in financial risk management and regulatory compliance.