Supervised Machine Learning Algorithms for Credit Card Fraudulent Transaction Detection: A Comparative survey

Abstract

Credit cards play a vital role in our day to day lives all over the world. Credit cards contain sensitive data and credit card frauds can affect not only the owner of the card but banks, government, and all type of financial organizations that causes in high financial losses. The numbers are expected to increase in the future, and for that reason, many researchers in this field have focused on detecting fraudulent behaviors early using various machine learning techniques. Basically, credit card fraudulent transaction detection becomes troublesome because of the two main reasons: one

of them is highly imbalanced data set and the other one is the fraudulent behav- iors usually differ for each attempt, so it is hard to find specific pattern/ features

for each fraudulent transaction. To get rid of these issues the algorithms like Lo- gistic Regression, Na ̈ıve Bayes, Random Forest, K- Nearest Neighbor, and Decision

tree (classification) algorithms can be used for detecting fraudulent transactions. A Comparative analysis is performed to find out which algorithm model performs best among them using different performance matrices like precision, recall, f1 score and provide an optimal solution. We have used data set from Kaggle, which is the highly imbalanced data set of credit card transactions of European credit cardholders of 2013. To balance the data set, Over Sampling, Under Sampling SMOTE (Synthetic Minority Oversampling Technique) techniques are used. Three re sampling techniques e.g. Over sampling, Under sampling and Smote are used to compare the results to know which one performs better. We have used Python programming language and JUPYTER NOTEBOOK IDE to perform comparison and view results.