Delay-driven Maximum Network Utilization by Machine Learning Model for Congestion Control

Abstract

The high speed growth of content-centric applications has drawn focus on efficient multi-source and multipath data transmission in Content-Centric Networking (CCN). The traditional approach for congestion control developed for stable, host-based architectures face several challenges due to CCN’s dynamic, cache-driven environment. It leads to inaccurate congestion estimation, bandwidth underutilization, and unstable fairness. To address these challenges, this article has proposed a Delay-Driven Congestion Control Protocol (DDCCP) that utilizes delay and other network parameters as input indicators for machine-learning-based decision models for precise congestion level estimation. The proposed work maintains autonomous control, enabling fair and effective bandwidth distribution without requirement of router-specific link-layer information. Five ML models—Decision Tree, k-NN, Naive Bayes, Fuzzy Logic, and SVM—are evaluated to guide the delay-based congestion adjustments, with SVM achieving the highest accuracy and lowest delay. Simulation results demonstrate that DDCCP significantly improves bandwidth utilization, reduces packet loss, accelerates fairness convergence, and adapts efficiently to caching effects in multipath CCN environments. The proposed framework offers a scalable, intelligent, and high-performance alternative to current CCN congestion control strategies.