Comparative Analysis of Machine Learning Models in Online Payment Fraud Prediction

With the substantial increase in Cybercrime, the pressing need for effective solutions in the light of the rising amount of corporate financial losses, is a concern f or many researchers. The widespread adoption of ecommerce and diverse online payment methods has contributed to the surge in a major form of cybercrime—online payment fraud. This further highlights the critical need to control and reduce the widespread danger of fraudulent behaviour on the internet. However, predicting online payment fraud is difficult as fraudsters are constantly changing their tactics, transactions happen quickly and in large volumes, transactions involve different regions and countries, stolen data is often used and other factors. Our goal is to improve the accuracy and usability of online payment fraud prediction through the use of machine learning. A well-established and widely recognized dataset specifically designed for studying online payment fraud was used. The six proposed algorithms trained and tested were (SVM), Decision Tree, Naïve Bayes, Random Forest, (KNN) and Logistic Regression. After rigorous testing, Decision Tree presented the highest accuracy in performance of 98.58%. However, as ten-fold cross validation was implemented on the dataset, Decision Tree suffered a loss while Random Forest had an increase in accuracy making it the best model with accuracy of 98.47%. As shown in this paper, machine learning has proven to be a reliable and ever-evolving approach for the prediction of Online Payment Fraud and results can be further improved in collaboration with financial and cybercrime specialists.