Comparative Case Study of Machine Learning Classification Techniques Using Imbalanced Credit Card Fraud Datasets

Today, the total transaction volume of credit cards is increasing consistently, as a result fraudulent transaction cases are also on a rise, producing losses in billions of dollars for financial institutions and banking sectors every year. Hence there is a need for a robust, reliable mechanism which is able to identify and prevent such fraudulent transactions effectively and efficiently. Some data mining techniques helps in detecting patterns between data attributes (classifying the transaction as fraudulent or non-fraudulent) and results in probabilistic prediction of the transaction category. In this study, multiple Machine Learning classification techniques are applied on a highly imbalanced datasets consisting of credit card transaction. ‘Chip and Pin' is considered as one of the trusted mechanisms today in terms of securing payment transaction but even this mechanism doesn't stops fake credit card utilizations on virtual Point Of Sale nodes or email orders known as an online 'credit card bankrupt'. It was observed that SVM, Random Forest and J48 Decision Tree classifiers yield a very high accuracy ratio but are suggested not to be leveraged while classifying such dataset where class imbalance is present. While thinking about these methodologies, this investigation gives a comprehensive overview of various classification methods, their highlights and restrictions of bankruptcy