ABSTRACT
E-commerce's explosive growth has made cloud computing even more crucial for scalable, real-time data analysis. This paper's hybrid anomaly detection approach addresses the challenges in identifying anomalous transactions in dynamic marketplaces by integrating deep learning, Bayesian Decision Networks, and Firefly-Simulated Annealing optimization. The framework combines two approaches: Bayesian modeling for probabilistic outlier detection and deep neural networks for the analysis of transaction patterns. Firefly-Simulated Annealing improves accuracy, reduces misclassifications by optimizing the detection thresholds, and processes large, unstructured datasets in real-time while adapting to transaction trends to distinguish between typical and anomalous activity. The system aims to make profiling more accurate, reduce false alarms, and offer real-time fraud detection that can scale for cloud-based e-commerce platforms. DL-BDN-Firefly surpasses traditional approaches with a 94% accuracy score, 92% in terms of precision and recall, and 91% scalability.This flexible framework enhances fraud prevention in e-commerce by offering a scalable solution with precision and real-time anomaly detection.
