Predictive Modeling for Fraudulent Credit Card Transactions

Currently, Credit card fraud is a big issue in the recent financial systems, causing huge financial losses on the consumer and the organization in the financial system. Thus, in order to eliminate the aspect of the same misuse and risk of financial loses by unauthorized transactions, good fraud detection systems should be very fast and very precise. To compare this several machine learning models such as Logistic Regression, Decision Trees, Random Forest, and XGBoost Naive Bayes, Random Forest does the best job among them due to the strong classification accuracy, rebalance capability against the rare class, and insensitiveness with respect to over fitting. The fraud detection workflow will be done in some steps which are Data acquisition, Preprocessing step, Feature creation step, Model construction, and last but not least Model deployment. As the fraudulent transactions count is greatly outnumbering the genuine transactions, it is highly recommended to balance the SMOTE, the dataset as such that the model can detect the fraud patterns very well. Furthermore, Feature scaling can also be made for the sake of the stability and dependability of the model. Then, the trained Random Forest model can be cast through Flask Web API which will be able to detect the fraud in real time in order to classify the transaction as a fraud or not at one spam time.