Advancing Financial Security with Scalable AI: Explainable Machine Learning Models for Transaction Fraud Detection

Background: The banking industry faces a critical challenge in maintaining data integrity while leveraging artificial intelligence (AI) for fraud detection, as financial fraud poses substantial risks to individuals and institutions. However, existing fraud detection systems often struggle with imbalanced datasets and lack of interpretability. Methods Used: This study employs machine learning (ML) techniques using the Financial Fraud Detection Dataset from Kaggle, incorporating data preprocessing, feature engineering, and class balancing. Models such as Random Forest (RF), AdaBoost, LightGBM (LGBM), and a Voting Classifier are trained and optimized using GridSearchCV for enhanced accuracy. Results Achieved: LGBM achieves the highest accuracy (90.20%), followed by the Voting Classifier (90.02%), while RF and AdaBoost record 89.26% and 88.37%, respectively. SHAP analysis provides insights into feature importance, enhancing model interpretability. Concluding Remarks: Financial institutions may rest easy knowing that explainable AI approaches provide openness and dependability due to the results showing that ensemble learning models work well for fraud detection.