Enhancing Online Fraud Detection: Leveraging Machine Learning and Behavioral Indicators for Improved Accuracy and Real-Time Detection

¹ School of Management & Research, Dr. D. Y. Patil Dnyan Prasad University's School of Management & Research, Pune, Maharashtra, India
² Surydatta Institute of Management and Mass Communication, Pune, Maharashtra, India

^* Corresponding author: shahaprasad@gmail.com

Published online: 18 June 2025

Abstract

Fraud detection remains a critical challenge in financial security, requiring robust and efficient methodologies to identify fraudulent transactions accurately. This study presents a comprehensive evaluation of machine learning (ML) models for fraud detection, emphasizing the role of behavioral indicators in enhancing model performance. A comparative analysis of traditional and advanced ML models, including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine (SVM), Artificial Neural Networks (ANN), and LightGBM, was conducted using real-world fraud detection datasets. LightGBM, the proposed model, outperformed other methods, achieving the highest ROC-AUC (0.981), F1-score (0.902), and lowest false positive rate (0.006). The study also highlights the importance of feature selection, class imbalance handling, and real-world applicability by discussing computational efficiency and deployment challenges. These findings contribute to the growing body of fraud detection research by offering a practical, scalable, and high-accuracy ML approach for real-time fraud prevention systems.