Paper Contents
Abstract
In modern network environments, static TCP congestion control algorithms (CCAs) such as Reno, Cubic, BBR, and Westwood fail to adapt to dynamic conditions, resulting in suboptimal throughput and latency. We propose a machine learning-driven framework that dynamically selects the optimal CCA in real time by monitoring network metrics (RTT, throughput, packet loss, bufferbloat, and retransmissions). Using a hybrid decision engine combining Random Forest and Long Short-Term Memory (LSTM) models with rule-based fallbacks, our system achieves an 88% prediction accuracy and reduces unnecessary CCA switches by 95%. Periodic model retraining with historical data stored in InfluxDB ensures adaptability. Experiments in namespace-based network simulations demonstrate up to 50% RTT reduction under bufferbloat and 20% throughput gain in lossy links. The lightweight design facilitates deployment in resource-constrained environments.
Copyright
Copyright © 2025 Suprith G B. This is an open access article distributed under the Creative Commons Attribution License.
