The evolution of credit card in financial transactions are massive throughout the world by generating millions of transactional records that have been maintained. These records have different patterns of genuine and fraudulent behaviour, the dataset is highly imbalanced with less fraudulent samples when compared with genuine. Machine learning trains the classifier based on the previous behavioural pattern and predicts the category of incoming transaction. In this study, we examined and compared the performance of four traditional supervised classifiers by applying raw data after removing few repeated features from the dataset. As the dataset is highly imbalanced, accuracy cannot be considered as a good metric. Therefore balanced classification rate was used to find the model accuracy. The efficiency of the classifiers was found by using different performance metrics. The results revealed that the fraud catching rate can be increased only when the dataset is balanced.
Volume 12 | 03-Special Issue
Pages: 1403-1409
DOI: 10.5373/JARDCS/V12SP3/20201391