An Ameliorated hybrid model for Fraud Detection based on Tree based algorithms and Benford's Law

In today's living conditions amidst pandemic the online dependence of buying and selling is at its peak. The world globally is looking for safer and convenient options of using online payment facility. The online sales are roaring because of pandemic conditions around us and the fraudsters are actively pursuing evolving strategies for fraud in online payment. The main aim of this paper is to develop a novel algorithm using machine learning to learn fraudulent patterns in credit card transactions and apply it on test data. The Benford's Law is used to find if deviations occur in data and hence determines if fraud transactions are present in dataset. The unsupervised K means algorithm produces output in the form of two clusters consisting of Fraud and genuine transactions. The predicted data along with unseen data reserved in beginning are passed to Logistic Regression, Random Forest and XG Boost. The credit card transactions are applied to hybrid model which detects whether transaction is fraud or not. The experimental results demonstrate the effectiveness of the proposed model. The proposed hybrid method gives excellent results in Roc -AUC score and Fl-score as compared to simple Random Forest and Logistic Regression.