A Hybrid Methodology for Detecting Anomalous Patterns of Imbalanced Data in Distributed Systems

Abstract

Securing distributed infrastructure is a highly complex challenge due to the sheer volume of data, the diversity of its sources, and the severe imbalances in its categories, which render traditional anomaly detection systems ineffective _{[1, 7]}. This work presents a hybrid approach combining privacy-preserving unified learning capabilities with local data balancing technology (SMOTE). This framework is designed to enable collaborative training across endpoints without requiring centralized raw data exchange, thus striking a balance between privacy and reliability _[9]. By applying heterogeneous data distribution processing locally at each site before global aggregation, this approach aims to minimize false positives and increase the likelihood of detecting attacks or rare anomalies _{[5, 12]}. This approach will help provide scalable, proactive protection for existing cloud computing and distributed networks _{[2, 8]}.

Keywords: Anomaly Detection, Federated Learning, Imbalanced Data, SMOTE Oversampling, Distributed Systems, Cyber security.