Fraud Detection in Banking Transactions Using Multi Layer Perceptron and Recursive Feature Elimination

Abstract

Fraud detection in financial transactions is a critical task for ensuring security and trust in banking systems. With the increasing volume and complexity of transactions, detecting fraudulent activities requires sophisticated machine learning techniques that can identify subtle patterns indicative of fraud. This study introduces a Multi-Layer Perceptron (MLP) augmented by Recursive Feature Elimination (RFE) fraud detection model for banking transactions. Finding fraudulent transactions based on important characteristics including transaction amount, customer information, payment method, and transaction time is the main goal. By removing redundant or unnecessary variables, RFE is used to choose the most pertinent features, greatly enhancing the model's performance. After using RFE, the accuracy increased significantly from 0.95 to 0.97 and Cohen's Kappa improved from 0.89 to 0.92. Furthermore, recall and precision rise, suggesting improved fraud transaction detection. The study shows that among the most important characteristics in fraud detection are transaction type, transaction value, and client age. Besides establishing the foundation for advanced feature selection research for the forthcoming trend in fraud, this paper emphasizes the importance of feature selection to improve model performance.