Machine Learning Classifiers for Detecting Credit Card Fraudulent Transactions

Abstract

Credit card usage has increased significantly as a result of the fast development of e-commerce and the Internet. As a consequence of enhanced credit card usage, credit card theft has risen substantially in recent years. Fraud in the financial sector is expected to have far-reaching effects in the near future. As a response, numerous scholars are concerned with financial fraud detection and prevention. In order to prevent bothering innocent consumers while detecting fraud, accuracy has become critical. We used hyperparameter optimization to see if created models utilizing different machine learning approaches are significantly the same or different, and if resampling strategies improve the suggested models’ performance. The hyperparameter is optimized using GridSearchCV techniques. To test the hypotheses of data that has been divided into training and test data, the GridSearchCV and random search methods are used. The maximum accuracy 72.1% was achieved by decision tree classifier on the imbalanced German credit card dataset. The maximum accuracy of 98.6% is achieved by LDA on imbalanced European credit card dataset. Additionally, logistic regression and naïve Bayes were also tested and SMOTE was applied.