Multi-Classifier and Deep Learning Framework for Robust Credit Card Fraud Detection

Credit card fraud remains a major concern in the financial sector, leading to significant financial losses and eroding customer trust. Real-time detection is challenging due to highly imbalanced datasets and constantly changing fraud behaviors. This work presents a unified fraud detection framework that combines supervised learning models such as Random Forest, Stochastic Gradient Descent Classifier, Support Vector Machine and Logistic Regression with unsupervised techniques including Isolation Forest, One-Class Support Vector Machine and Density-Based Spatial Clustering for Applications with Noise to identify both known and previously unseen fraud patterns. Reinforcement learning methods such as Q-Learning and Deep Q-Learning enable adaptive detection strategies that improve over time, while deep learning architectures such as Multilayer Perceptron and Long Short-Term Memory networks capture complex sequential and nonlinear transaction patterns. A key contribution is the extraction and conversion of model parameters into FPGA compatible formats using Vitis High-Level Synthesis, enabling real-time hardware-accelerated inference with significantly reduced latency. Experimental evaluation on benchmark datasets demonstrates superior accuracy, recall, F1-score and ROC-AUC with lower false-positive rates compared to individual models, making the proposed framework scalable, adaptive and effective for real-time credit card fraud detection..