Credit Card Fraud Detection Using Machine Learning, Deep Learning, and Ensemble of the both

It is the responsibility of the credit card companies to ensure that their customers are given the best security measures when it comes to fraudulent credit card transactions. The main objective of this work is to explore whether a credit card transaction (before being processed) is fraudulent or not. And if it is found out to be fraudulent, then the credit card owner must be notified of the same. The approach used is to make use of highly imbalanced and skewed transactional data and train various models available in ML as well as a DL model for the detection of frauds and finally compare their accuracies and test results and choose the most suitable model. The system currently used to detect fraud is plagued by misclassifications and highly false positives. So, our motive is to not miss any fraud cases (high recall) as well as not predict too many non-fraud cases as fraud cases (high precision) as if this happens, it will lead to poor reviews for the concerned credit card company. We demonstrate various methods to deal with the imbalance of the data such as choosing appropriate metrics for evaluation of models and resampling of data.