Anomaly Detection Model in Online Transactions Using Supervised Learning Techniques

The increasing prevalence and sophistication of credit card fraud pose significant security challenges to financial institutions and individual users worldwide. As fraudsters continuously adopt advanced techniques to exploit vulnerabilities in online transaction systems, there is a growing need for intelligent and reliable fraud detection mechanisms. This paper investigates the effectiveness of a Feed Forward Back Propagation Neural Network (BPNN) in modeling anomaly detection for online credit card transactions.

The study focuses exclusively on online credit card transactions and employs a large-scale dataset consisting of 284,807 transaction records obtained from Kaggle. These records include both demographic and transaction-related variables. The proposed fraud detection model was developed and simulated using MATLAB, with 70% of the dataset allocated for training and 30% reserved for validation through testing.

Experimental results demonstrate that the Feed Forward BPNN achieves a high classifier accuracy of 99.9%, an Area Under the Precision-Recall Curve (AUPRC) of 59.3%, and an overall prediction accuracy of 79.9%. These findings confirm the capability of the BPNN model to effectively identify fraudulent activities in online credit card transactions. The study therefore highlights the potential of neural network-based machine learning techniques as robust tools for enhancing transaction security and mitigating financial fraud in online environments.