AUTONOMOUS NETWORK MONITORING USING COLLABORATIVE LEARNING FOR DISTRIBUTED TRAFFIC CLASSIFICATION

Abstract

Conventional traffic monitoring methods are becoming less efficient as numerous applications are rapidly adapting to counteract attempts to identify them, which creates new challenges for traffic monitoring. Autonomous Distributed Network Monitoring (ADNM) scheme is a promising approach to address these challenges, nonetheless each ADNM node has its own limitation, such as adapting to concept drift and self-learning, hence needs to collaborate with other autonomic nodes to monitor the network efficiently. This paper, presents a collaborative learning and sharing structure among self-managed network monitoring nodes, ensuring interaction for efficient information exchange for distributed autonomic monitoring towards the achievement of global network management objectives. A machine learning algorithm for collaborative learning among distributed autonomic monitoring nodes is proposed. This algorithm is based on the concept of online incremental k-means traffic classification model. Experimental results using publicly available real network traffic traces shows that network nodes participating in collaborative learning performs better with a higher overall total average accuracy of about 4.5% over centralized monitoring nodes. The overall performance indicates that collaborative learning is a promising technique that can value-add to the overall distributed network monitoring performance