Credit card fraud detection is one of the essential tasks of the banking system. However, there are various Machine Learning (ML) solutions available for detecting and preventing potential fraud. In this paper, the aim is to compare the recently contributed ML-based credit card fraud detection technique. In this context, three different models are considered, first a brief introduction of the considered approaches has been given. Then, based on experimental results the comparison among the techniques has been discussed. Based on the findings, the publically available credit card fraud detection dataset has suffered from the class imbalance problem, large dimensions, missing values, overlapped attributes, and outliers. Therefore, two key approaches are discussed first is based on handling the class imbalance problem and the second is based on deep cleaning of the dataset. According to the results, we found that the deep clean method is superior then the over-sampling technique. The deep clean technique reduces the dimensions and also removes the noise from the dataset. Therefore the method reduces the data processing cost in terms of time and memory. Additionally, the classification accuracy has a similar behavior as the oversampling. In both cases, the performance varies between 97.9%- 99.2%. Finally, based on the comparative study the conclusion of the work has been made.

Machine Learning, Classification, Credit card fraud detection, machine learning application, supervised and unsupervised learning, comparison.