Scholarlink

<p>With the number of cyber finance fraud cases increasing dramatically, posing a challenge to social and economic development. Rapid and accurate identification of potential cyber finance fraud can effectively prevent these crimes from occurring. This paper employs ensemble machine learning techniques to analyze abnormal transaction behavior data, aiming to uncover the characteristics of cyber finance fraud facilitation behaviors. It specifically focuses on clarifying the key types of cybercrime facilitation and constructing a robust model to identify these malicious activities. Based on case information from a financial investigation department in China and original bank transaction data, the dataset includes comprehensive information from individuals, banks, and regulatory agencies, enabling an in-depth analysis of cybercrime facilitation behaviors. After evaluating a range of algorithms, including Naive Bayes, SVM, MLP, Random Forests, LightGBM, and XGBoost, we found that SMOTE-XGBoost provided superior performance. The accuracy of the SMOTE-XGBoost model increased from 90.17% to 92.25%, highlighting its enhanced classification capability, while the recall rate rose from 90.17% to 93.05%, indicating improved sensitivity in detecting actual fraudulent transactions and effectively capturing more fraud cases. Additionally, by employing the SHAP (SHapley Additive exPlanations) interpretable machine learning method, we can quantify the influence of synthetic samples and various feature data on model decisions. This approach enables effective global and local explanations of the model, thereby increasing its credibility and transparency. The model results reveal that the number of associated counterparty accounts, account age, and transaction volume are key indicators that play a crucial role in effectively identifying abnormal trading behavior. Therefore, combining SMOTE-XGBoost with SHAP facilitates intelligent detection of behaviors associated with cybercrime and offers a novel approach for uncovering potential cyber fraud schemes.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Cyber Finance Fraud Recognition Method Based on Ensemble Machine Learning

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Jiguang Shi,
  • Shancheng Lin,
  • Ning Ding,
  • Jianfeng Song,
  • Yan Zhai

摘要

With the number of cyber finance fraud cases increasing dramatically, posing a challenge to social and economic development. Rapid and accurate identification of potential cyber finance fraud can effectively prevent these crimes from occurring. This paper employs ensemble machine learning techniques to analyze abnormal transaction behavior data, aiming to uncover the characteristics of cyber finance fraud facilitation behaviors. It specifically focuses on clarifying the key types of cybercrime facilitation and constructing a robust model to identify these malicious activities. Based on case information from a financial investigation department in China and original bank transaction data, the dataset includes comprehensive information from individuals, banks, and regulatory agencies, enabling an in-depth analysis of cybercrime facilitation behaviors. After evaluating a range of algorithms, including Naive Bayes, SVM, MLP, Random Forests, LightGBM, and XGBoost, we found that SMOTE-XGBoost provided superior performance. The accuracy of the SMOTE-XGBoost model increased from 90.17% to 92.25%, highlighting its enhanced classification capability, while the recall rate rose from 90.17% to 93.05%, indicating improved sensitivity in detecting actual fraudulent transactions and effectively capturing more fraud cases. Additionally, by employing the SHAP (SHapley Additive exPlanations) interpretable machine learning method, we can quantify the influence of synthetic samples and various feature data on model decisions. This approach enables effective global and local explanations of the model, thereby increasing its credibility and transparency. The model results reveal that the number of associated counterparty accounts, account age, and transaction volume are key indicators that play a crucial role in effectively identifying abnormal trading behavior. Therefore, combining SMOTE-XGBoost with SHAP facilitates intelligent detection of behaviors associated with cybercrime and offers a novel approach for uncovering potential cyber fraud schemes.