Derin öğrenme tabanli yöntemler kullanilarak e-ticaret platformlarinda anomali kullanici davranişlarinin tespiti
Detection of Anomalous User Behaviors in E-Commerce Platforms Using Deep Learning–Based Methods

Summary/Abstract: With the rapid growth of e-commerce platforms, detecting anomalous transactions and user behaviors from large-scale operational data has become a critical requirement for ensuring system security and operational efficiency. This study presents a comparative analysis of several deep learning– and machine learning–based anomaly detection approaches applied to e-commerce systems. The dataset used in this research was obtained from the Kaggle platform and consists of 17,049 observations with 18 attributes, including customer demographic information, order and payment details, delivery time, website interaction metrics, and customer satisfaction scores. This rich combination of behavioral and transactional features provides a realistic representation of real-world e-commerce environments. The experimental evaluation includes five anomaly detection methods: Autoencoder, LSTMAutoencoder, Deep Support Vector Data Description (Deep SVDD), a GAN-based reconstruction approach (GAR-AD), and Isolation Forest. Model performance was assessed using AUROC, AUPRC, best F1 score, and Top-K metrics (Top 1%, 5%, and 10%) to reflect practical investigation capacity constraints in real-world e-commerce operations. The results indicate that the Autoencoder model achieves strong discriminatory performance, while the GAN-based GAR-AD approach provides more balanced and robust results in terms of AUPRC and best F1 score. In particular, the Top-K analysis demonstrates that GAR-AD yields superior precision and lift values under limited inspection budgets.

• Details
• Contents