A Stacked LightGBM-XGBoost Model with SHAP-Based Fraud Detection for Financial Transactions

Financial transaction fraud detection is a significant task that requires robust and efficient machine learning algorithms to be deployed. In this research, we proposed a hybrid model that uses Light Gradient Boosting Machine (LGBM) and Extreme Gradient Boosting (XGB) to enhance the accuracy of fraud detection. We employ diverse feature engineering techniques, preprocessing, and model performance measures to enhance the classification of fraudulent behavior. The suggested model is trained on a transaction and identity data set, which performs better in terms of accuracy and Area Under the Curve (AUC) value. Additionally, the model is also tested using SHAP (Shapely Additive Explanations) values, which provide more precise insights into feature importance. We also highlight the importance of Artificial Intelligence (AI) integration through the application of gradient boosting algorithms to the enhancement of feature cybersecurity in fraud detection through improved predictive accuracy and reduced false positives. Keywords— AI enhancement, Fraud Detection, LightGBM, XGBoost, Machine Learning, Financial Security, Gradient Boosting, SHAP.