INTELLIGENT TRAFFIC MANAGEMENT IN SOFTWARE-DEFINED NETWORKS THROUGH DATA-DRIVEN MULTI-AGENT REINFORCEMENT LEARNING

Abstract

Software-Defined Networks (SDNs) have revolutionized network management by separating the control and data planes, enabling centralized traffic control. In India, rapid growth in internet users over 1.2 billion mobile subscriptions and rising data traffic has led to increasing congestion and packet loss in networks. Traditional approaches struggle with dynamic traffic patterns. The objective is to enhance congestion detection and traffic control in SDNs using machine learning. It enables intelligent, adaptive network management, reducing delays and improving throughput dynamically. Manual network traffic control relies on human monitoring and static routing rules. Network administrators adjust traffic paths and priorities based on observed congestion patterns and historical usage data. Tools like SNMP and simple logging systems provide basic insights but require constant human intervention to optimize performance. Manual systems are slow, error-prone, and inflexible. They cannot respond to real-time traffic fluctuations, often resulting in packet loss, increased latency, and inefficient bandwidth utilization. Scalability is limited, making them unsuitable for large, high-speed SDN environments. The motivation is to overcome manual limitations by leveraging intelligent, adaptive methods. Machine learning models like SVM, KNN, and TaoTree address inefficiency and latency issues, providing real-time congestion detection and traffic prediction. The system improves reliability, accuracy, and scalability, supporting dynamic network management for high-traffic SDN environments. The proposed system uses data-driven machine learning models for adaptive congestion detection and traffic control in SDNs. SVM handles high-dimensional traffic data, KNN predicts traffic patterns using neighbor instances, and TaoTree with SMOTE balances multiclass datasets for accurate classification. These models analyze real-time network traffic, detect congestion proactively, and predict throughput and delay. By integrating predictions into SDN controllers, the system dynamically adjusts routing and bandwidth allocation, improving network efficiency, reducing packet loss, and enabling self-optimizing traffic management