ONLINE FRAUD PAYMENTDETECTION USINGBALANCEDML ALGORITHMS

Abstract

Online financial transactions have increased rapidly in recent years, leading to a significant rise in fraudulent activities. Fraudulent payments can result in severe financial loss, compromised customer trust, and disrupted business operations. A major challenge in fraud detection is the highly imbalanced nature of payment datasets, where legitimate transactions significantly outnumber fraudulent ones. Traditional machine learning models often fail to detect minority-class fraud samples accurately, resulting in high false-negative rates. This paper proposes a fraud detection system using balanced machine learning algorithms combined with sampling techniques such as SMOTE, Random Undersampling, and Hybrid Sampling. The system uses transaction features including amount, time, device ID, location, and user behavior patterns. Balanced classifiers like Random Forest, XGBoost, and Logistic Regression with class weighting are employed to enhance fraud detection accuracy. Experimental results demonstrate that using balanced datasets leads to significantly improved recall and F1-score, ensuring more reliable fraud identification. This approach minimizes false negatives, enhancing security in digital payments. The proposed model is scalable, adaptable, and suitable for implementation in real-world financial systems.