Hybrid Lightweight Anomaly Detection for Credit Card Fraud

Credit card fraud has become a major challenge in modern digital payment systems,
requiring fast, accurate, and scalable fraud detection solutions. This paper proposes a Hy-
brid Lightweight Anomaly Detection System designed to identify fraudulent transactions
in real time with minimal computational overhead. The framework integrates statistical Z-
Score filtering, Isolation Forest–based unsupervised anomaly scoring, and Logistic Regres-
sion for interpretable fraud probability estimation. The system processes transaction data
through optimized preprocessing, feature engineering, and hybrid scoring to capture both
rare outliers and subtle fraudulent patterns. Experimental evaluation using a benchmark
credit card fraud dataset demonstrates that the proposed method achieves high accuracy
(96.1strong recall for fraud cases, and extremely low inference time (¡0.02 seconds per
transaction), making it suitable for real-time fintech applications. The lightweight design
ensures easy deployment on low-resource environments while maintaining high reliability
and transparency. The proposed system enhances financial security by reducing false pos-
itives, improving early fraud detection, and offering a scalable solution for modern digital
payment ecosystems
Keywords:Credit Card Fraud Detection, Hybrid Anomaly Detection, Li