Fraud Detection in Banking Transactions Using...

Fraud Detection in Banking Transactions Using Machine Learning

Author Name : N. Manmadha Rao, Mr. Md Imam Khader Shariff, R. Trilokya, L. Suresh

DOI: https://doi.org/10.56025/IJARESM.2026.140426

ABSTRACT Financial fraud causes billions in annual losses, yet traditional rule-based systems struggle with evolving patterns and severe class imbalance. This paper presents an XGBoost-based fraud detection system using the PaySim synthetic dataset (6.36M transactions, 0.13% fraud rate). We engineer novel features—balance error, sender/receiver amount ratios, and outlier capping—to expose fraudulent patterns invisible to standard features. SMOTE addresses class imbalance, and we compare XGBoost against Random Forest and Logistic Regression using F1-score, ROC- AUC, and PR-AUC on imbalanced test data. XGBoost achieves 99.95% accuracy, 0.9999 AUC, and optimal F1 at threshold 0.3. A novel contribution is practical deployment via Streamlit web app supporting single-transaction input and CSV batch prediction—ready for banking integration. Results validate feature engineering and gradient boosting for real-world fraud detection in highly imbalanced scenarios.