TrustNet: Deep Learning Approach for Securing Internet-Based Loans

Abstract

The rapid expansion of internet-based loan services has exposed financial institutions to sophisticated fraudulent activities. Online loan systems are highly vulnerable due to the absence of direct human verification, allowing fraudsters to exploit loopholes with false information. This paper proposes TrustNet, an intelligent fraud detection system using Artificial Neural Networks (ANN) to identify fraudulent loan applications. The system analyzes applicant details including income, credit history, employment status, and loan attributes. Essential preprocessing techniques including missing value handling, categorical encoding, and feature normalization enhance prediction performance. The model is deployed via a Flask web interface for real-time fraud detection. Evaluation on a historical loan dataset of 15,000 applications demonstrates 95.2% accuracy with 0.94 F1-score, outperforming traditional machine learning approaches (Random Forest: 88.6%, Logistic Regression: 82.3%) in detecting complex fraud patterns