A Comparative Analysis of Machine Learning Algorithms on Card-Based Financial Fraud Detection with Infusion of Sigmoid and Isotonic Functions

Abstract

There is no doubt that the commencement of the e-revolution in the financial sector of the economy has introduced opportunity for electronic fraud in the card payment ecosystem globally. This is occasioned by factors such as increased knowledge in the fintech space, poverty, pair pressure on the perpetrators. This study was focused on comparing of several known Machine leaning Algorithms – Logistic regression, Decision trees, Random Forest, Extra-Trees, Adaboost and Gradient boosting on how they perform comparatively when applied to Fraud detection. The raw data used were obtained from The Xente Fraud Detection data set used for the development of the financial fraud detection model which includes sample of approximately 140,000 transactions categorized into Fraud and Non-Fraud. Results from the study indicated that Adaboost Model outperformed the remaining applied models thereby making Adaboost a good model for fraud detection. Further research work can be carried out by comparing the performance with other Deep Learning Algorithms.