Enhancing Fraud Detection in Online E-Commerce Transactions through Deep Learning Auto encoder Model

Fraud detection in internet transactions is crucial for maintaining security in digital transactions. Traditional methods often rely on rule-based systems or supervised learning algorithms, which struggle to capture complex patterns in fraudulent behaviour and require extensive manual feature engineering. In this study, a novel approach using deep learning auto encoder models is proposed to enhance fraud detection accuracy. Unlike conventional methods, this approach does not necessitate fraudulent data for training, as the auto encoder learns to reconstruct legitimate transactions and identify anomalies based on reconstruction error. The trained model is integrated into the market for real-time detection. Results demonstrate significant improvements in fraud detection accuracy compared to traditional methods, with reduced reliance on manual feature engineering and labelled fraudulent data. This approach offers a scalable and effective solution for detecting fraudulent activities in online e-commerce transactions, thereby safeguarding users and businesses from financial losses and maintaining trust in digital transactions. The autoencoder model expands detection accuracy by 2% over RF and 1% over SVM, with an AUCROC increase of 1%.