Paper Key : IRJ************833
Author: Kamayoyo Mulele Mufuzi,Charles Lubobya,Smita Francis
Date Published: 05 Apr 2024
Abstract
Abstract - Border Gateway Protocol (BGP) is susceptible to anomalies severely degrading network availability and performance. However, accurately detecting unknown anomalies in dynamic communication networks remains challenging. This research implements and evaluates Long Short-Term Memory (LSTM) networks for anomaly detection in evolving BGP networks. Public BGP datasets containing routing updates were used to train and test the LSTM models. Key performance metrics like accuracy, false positives, ROC curves, and overhead were measured and analysed. LSTM achieved 93% accuracy in detecting route hijacks and outperformed baseline classical machine learning algorithms. However, it incurred substantially higher overhead during training and inference versus simpler models. LSTM delivered strong capabilities for BGP anomaly detection amidst concept drift.Further operational testing would refine the approach. This work provides empirical insights into deep learnings potential while outlining real-world feasibility constraints. Effectively securing critical infrastructure against emerging threats requires adaptive, efficient anomaly detection.Keywords: Anomaly detection, Long Short-Term Memory (LSTM), Border Gateway Protocol (BGP), Deep learning, Cybersecurity