NEW BGP ROUTE LEAKS CLASSIFICATION AND DETECTION USING SUPERVISED MACHINE LEARNING TECHNIQUE

Abstract

The route leaks problem is considered one of the unsolved problems in BGP through the previous 15 years. The confidentiality of autonomous systems (ASes) relationships and the lake of advertisement of route leaks incidents are the main two reasons behind this. This thesis solves the route leaks problem relaying on three steps: A new taxonomy for the route leaks types based on their effects to the BGP Traffic not to their ASes relationships is proposed, the first dataset for published real route leaks incidents through the previous years is collected and labeled as route leaks or normal traffic, and the first real-time detection system of route leaks problem using complex features extracted from BGP Update messages only and using Classification algorithms is proposed. The system achieves the best accuracy of 88% and 92% F1 Score, the whole system can detect route leaks upon receiving them in real-time as it runs in less than one second.