This project deals with building an end-to-end model to detect the occurrence of fraudulent transactions in a credit card company.
Dataset contains around 284,807 transactions with only 492 fraud transactions suggesting high imbalance in the datatset
-
Data Preprocessing: Data was scaled and Principal Component Analysis was applied to the data to reduce dimensionality while still retaining maximum information
-
Model Training: Multiple models like Decision Tree, KNN, Logistic Regression, SVM, Random Forest, XGBoost have been trained on the transformed dataset
-
Model Evaluation: Keeping in mind the high imbalance in the dataset, F1 score has been used to evaluate the peformance of the model
-
Improvisation: To further improve the performance of the models, hyperparameter tuning was conducted to obtain the best set of paramaters for each model
-
Conclusion: Finally, the performance of all models was compared to reveal the best performing model