ML-Assisted Dynamic Multi-Path Routing for Enhanced QoS

The increasing reliance on the Internet for various applications has intensified the need for robust Quality of Service (QoS) in network traffic management. Conventional single-path routing methods frequently fall short of meeting contemporary network applications' high bandwidth and low latency demands. This paper introduces a QoS-aware dynamic routing framework for edge routers that leverages machine learning (ML) methods to classify network packets and employs a multi-path routing approach to adapt dynamically to changing network conditions. The proposed framework comprises three main modules: a packet classifier, a load balancer, and a routing engine. The packet classifier classifies the internet packets using ML. The load balancer marks flows based on priority levels and analyzes link loads to prevent congestion. The routing engine then selects the optimal paths based on these analyses, ensuring efficient data transfer with optimization routing cost. In the evaluation part, the packet classifier uses pre-trained ensemble learning methods. According to classification results, we obtain 0.971, 0.950, 0.914, and 0.719 Fl scores for XGBoost, Random Forest, LightGBM, and Catboost, respectively. In the routing part, we conduct our test using NS-3 network simulator. We compare our method with Nix-Vector Routing Protocol and Open Shortest Path First (OSPF) regarding packet delivery ratio and throughput. Our multi-path with XGBoost method received the highest packet delivery ratio throughout the simulation, starting with 95% and declining to 90%. Also, multi-path with XGBoost achives 50% and 38.46% better performance in terms of throughput than Nix-Vector and OSPF, respectively.