This project presents a hybrid machine learning framework designed to detect credit card anomalies and fraud patterns using a multi-stage processing approach. The system integrates several learning techniques, including Convolutional Neural Networks, Stacking, Voting-based ensembles, XGBoost, and TabNet. The framework first prepares the highly imbalanced dataset through structured preprocessing steps, followed by feature learning and classification. Each model contributes unique strengths: CNN captures local patterns, XGBoost handles complex interactions, TabNet performs attentive feature selection, and ensemble methods enhance stability. The system is deployed using a Flask backend with an interactive interface built with HTML, CSS, and JavaScript. The platform includes modules such as Home, Register, Login, Classification, Model Performance, and Logout. The goal is to support accurate, fast, and interpretable detection of unusual credit card behavior. Experimental evaluation shows that the hybrid approach provides consistent improvements over individual models. The framework demonstrates an efficient and flexible design suitable for anomaly detection settings where precision and reliability are essential.

Keywords: credit-card-fraud, anomaly-detection, CNN, XGBoost, TabNet, Stacking, Voting, Flask, classification, imbalanced-data.