Abstract:
This thesis focuses on financial fraud detection, a major challenge for modern banking systems.
The goal is to build an efficient machine learning model that can identify fraudulent transactions in
imbalanced datasets.
Real-world payment data was preprocessed and analyzed using supervised models such as Decision
Trees and Random Forests. Multi-objective optimization with the NSGA-II algorithm was applied to
enhance the balance between precision and recall.
Results show that the optimized Random Forest model achieves 78% recall on the validation set
while keeping false positives low. This work contributes to improving automated fraud detection
systems and lays the groundwork for future hybrid or real-time approaches.
