Machine Learning for Adaptive RealTime Fraud Detection in Financial Systems.

Fraud identification and credit risk evaluation are essential components of financial data analysis systems. This project introduces a machine learning–based framework designed to detect multiple fraud categories and classify credit risk using a synthetic dataset. The proposed system performs multi-class classification to identify phishing, card fraud, payment fraud, lottery fraud, and non-fraud cases, while also conducting binary classification to determine low-risk and high-risk credit profiles. To improve predictive performance and model stability, ensemble learning algorithms such as Random Forest, XGBoost, and LightGBM are implemented and compared. Class imbalance within the dataset is addressed using the Synthetic Minority Over-sampling Technique, which enhances minority class representation during training. In addition to accuracy, the project emphasizes model interpretability through explainable AI techniques that provide insight into feature contributions influencing predictions. The system is deployed through a Flask-based backend integrated with a web interface that allows users to upload datasets, explore data distributions, and obtain prediction results. The modular design supports clear separation between data processing, model execution, and user interaction. This project demonstrates a comprehensive analytical workflow that combines data preprocessing, advanced classification models, and interpretability to support research-focused evaluation of fraud patterns and credit risk behavior using structured synthetic data.

Keywords: Fraud Detection, Credit Risk Classification, Machine Learning, Ensemble Models, SMOTE, Explainable AI, Synthetic Dataset, Flask Framework, Multiclass Analysis, Risk Prediction.