AN ENHANCED FEATURE ENGINEERING TECHNIQUE FOR CREDIT CARD FRAUD DETECTION

Abstract

As the world is becoming a cashless society with increasing use of online transactions, the number of credit cards users has also increased substantially. This led to credit card fraud, which is among the major cybercrimes faced by users with consequential damages to financial institutions. Therefore, credit card fraud detection is crucial due to the increasing number of credit card transactions. Machine learning based credit card fraud detection systems exist, but machine learning approaches have problems with imbalanced data and the need to selected best features for effective classification. Imbalance classification occurs when there are small number of observations of the minority class compared with the majority in a dataset. This study addresses the challenges of feature selection and data imbalance in credit card fraud detection through an enhanced feature engineering method. We propose a technique that uses wrapper to select the best features and mitigate data imbalance using a hybrid approach that combines SMOTE, random oversampling and under-sampling techniques. Five popular machine learning classifiers—Random Forest, Naïve Bayes, K Nearest Neighbor, Decision Tree and Support Vector Machine—are used with balanced and imbalanced datasets to evaluate the technique. The results show significant improvements in accuracy, precision, recall, F1-score, and Kappa score with the enhanced method. Specifically, and K Nearest Neighbor, Random Forest and Support Vector Machine achieve perfect accuracy with the balanced data.