Enhancing Trust in Digital Payments: Benchmarking Machine Learning Models for Transactional Fraud Detection

Abstract: The exponential growth of digital banking has heightened transactional fraud risks, resulting in significant financial losses. This study introduces a real-time fraud detection system employing an ensemble of Statistical Processing Model (SPM), K-Nearest Neighbors (KNN), Logistic Regression, and Convolutional Neural Networks (CNN) to monitor user transactions characterized by amount, geolocation, device fingerprint, IP address, frequency patterns, and behavioral history. KNN detects anomalies against user-specific baselines, Logistic Regression computes fraud probabilities, and CNN extracts deep spatiotemporal features from sequential transaction data to identify complex fraud signatures. Detected anomalies trigger immediate security responses including user notifications, account suspension, and administrative alerts. Evaluation demonstrates superior AUC-ROC and F1-scores compared to baseline methods, validating the system's efficacy for scalable, production-ready deployment in securing digital payment ecosystems while preserving legitimate user experience.

Keywords: transactional fraud detection, digital banking security, ensemble machine learning, CNN anomaly detection, real-time fraud prevention, behavioral biometrics.