Online Payment Fraud Detection using Machine Learning

Abstract

The rapid growth of e-commerce has greatly boosted the frequency of online payment fraud, and poses a great challenge to both financial institutions and consumers. This paper proposes a machine learning-based fraud detection system utilizing random forest and decision structure algorithms, which are capable of detecting suspicious transactions efficiently. This model is trained according to historical transaction data and has characteristics such as steps, transaction type, Amount, oldbanance(origin), newbalance(origin),oldbanance(dest),newbanance(dest) and fraud indicators. Comprehensive testing shows strong accuracy, accuracy, recall and F1 scores for the model, which reduces both false positives and false negatives.
This work also examines how selecting and adjusting characteristics via hyperparameters affects and reports that fine-tuning of such factors improves the effectiveness of identification. To compensate for the lessons that characterize fraud perception, this study uses class weighting techniques to improve model skills and recognize fraudulent transactions. The results examine the potential for machine learning to improve fraud prevention systems and provide a scalable, customizable solution to improve the security of digital financial transactions.