Project Goal

Achieving the best balance between detection of fraud and non-fraud transactions in an imbalanced dataset

Data Preprocessing

Using SMOTE technique to make the dataset balanced

Models

• Logistic Regression
• Random Forest Classifier
• XGBoost Classifier
• Neural Network

Conclusions

1. Good prediction results can be achieved with imbalanced datasets as well as with balanced ones
2. Random Forest and XGBoost Classifiers gave us the best results being able to detect more than 80% fraud transactions and at the same time not classifying a lot of non-fraud transactions as fraud
3. There is no perfect model and there will always be a trade-off between precision and recall. It is up to the company and its objectives to decide which approach is the best in each particular situation