This project investigates the efficacy of various machine learning algorithms at detecting fraud in three different areas of financial activity: Credit Card Transactions, Ethereum Network Transactions, and Insurance Claims. We use the FLAML (A Fast Library for Automated Machine Learning & Tuning) python package for training, tuning, and evaluating the performance of various different classes of machine learning models. We find that the XGBoost and LGBM FLAML techniques consistently outperform other methods across all three areas of financial activity. Our results offer direction for future, more rigorous investigations into machine learning techniques in different transactional contexts and benefit financial firms, individuals, and researchers that have an interest in machine learning as it applies to transactional data.
The primary conclusion from our project is that the Xgboost and LGBM FLAML algorithms perform best in all three areas of financial activity and that algorithm performance is largely not dependent on the area of financial activity. Conclusions for each domain specific goal for each of the three datasets can be found of our github. We intend for these results to be useful for anyone interested in implementing FLAML fraud detection algorithms and hope to see our research used to help deploy effective models.
Each team member was responsible for the preprocessing and preparation of their own dataset for both their individual goals and the team goal. Ben primarily implemented the algorithms and analyzed the results. Each of the three team members contributed to the project paper. Ben and Jie primarily created the final presentation.