Securing E-Commerce Platforms using Log BERT: A Self-Supervised Transformer-Based Anomaly Detection Approach in Cloud Environments
Universitas Andalas Padang, West Sumatra, Indonesia, Farid Hidayat · International Journal of Multidisciplinary Research in Science, Engineering and Technology. · 2023
The ever-increasing growth of e-commerce platforms has created a heightened demand for well, scalable, intelligent cybersecurity solutions for protecting sensitive data and maintaining system integrity. Traditional anomaly detection mechanisms, being mostly rule-based and so supervised, have been inefficient in dynamic cloud environments, generally failing to generalize unseen patterns; they possess high false alarm rates since much of their operation relies on labelled datasets. These weaknesses are fully addressed in this article through the presentation of a model for self-supervised transformer-based anomaly detection-Log BERT-targeted for multi-cloud e-commerce environments. This framework utilizes transformer architecture to model complex, temporal dependencies inherent in unstructured log data without requiring any form of labelling of the input data. Log sequences are generated through session-based sliding windows and enriched with sinusoidal positional encodings; this masked language modelling (MLM) serves as the pretext task required for the model to learn contextual representations. Anomalies are eventually scored using negative log-likelihood and further classified via One-class Support Vector Machine (OC-SVM) for high- precision outlier detection. The entire detection pipeline is securely deployed in the cloud using AES-256 encryption in conjunction with a Cloud Key Management Service (KMS), thereby ensuring compliance with data protection standards. The proposed system shows considerable performance improvements in inference latency, detection accuracy and storage efficiency at decreased dependence on manual rule creation or data labelling. This is a highly intelligent and secure scalable solution to a major problem in modernizing the security of an e-commerce infrastructure against evolving threats faced in distributed cloud-native settings.