Enhancing Network Management using Machine Learning: Intent Based Networking

Abstract

IBN which is intent-based networking, is changing the way network administrators manage networks with automated intelligent processes. In this paper, we propose a machine learning (ML)-enabled IBN system with automatic programming, intelligent intent translation, adaptive optimization and proactive anomaly detection. With use of cutting-edge technologies like NLP (natural language processing), neural networks, deep-learning and RL(reinforcement learning) our system efficiently translates high-level business requirements in low level network configurations during runtime. We performed a thorough evaluation of the system in a simulated business network environment and compared it with traditional IBN solutions (see, e.g., [3, 4]) as well as current works. The results were highly encouraging: our system achieved 96.56% translation accuracy, improving the state of the art. It also reduced the network convergence by 73.4%, and was much faster to adapt to network changes. We showed that the anomaly detection module attained high performance with F1 ¼ 0:9485 and thus a notable precision and recall. Resource utilization also attained an average gain of 29.05% as compared to SoA approaches. Scalability tests proved that the system performed consistently as network size and complexity increased, a task many large-enterprise systems struggle with. Such results illustrate the benefits of ML-based IBN for better efficiency and automation in network operation. Our work contributes towards the emerging field of intelligent network management with practical applications for both researchers and practitioners. This work will help shape the next generation of IBN which would focus on more self-reliant, agile and operationally efficient infrastructure.