INTELLIGENT CREDIT SENTINEL: A SYNERGISTIC MULTI-LAYERED MACHINE-LEARNING FRAMEWORK FOR DETECTING FRAUD AND BILLING ANOMALIES

ABSTRACT Traditional monolithic systems struggle against the dynamic nature of financial fraud and other transactional risks. This paper introduces the Intelligent Credit Sentinel, a novel four-layered hierarchical architecture designed for robust and multi-faceted transaction risk assessment. The system employs an unsupervised autoencoder for broad anomaly detection (Layer 1), two specialized supervised XGBoost classifiers for targeted fraud and billing error detection (Layers 2 & 3), and a final logistic regression meta-learner to synthesize these outputs into a single, actionable risk score (Layer 4). Through extensive feature engineering and hyperparameter tuning to manage severe class imbalance, the specialist layers achieved high performance, with the final billing error model attaining 94% precision. The synergistic combination of these layers in the meta-learner resulted in a final system-wide recall of 82.4% for all high-risk events. The findings demonstrate that this modular, tiered approach is a highly effective, interpretable, and operationally efficient paradigm for modern financial security. KEYWORDS: Fraud Detection, Multi-Layered Architecture, Meta-Learner, Anomaly Detection, XGBoost, Class Imbalance.