Congestion Prediction using Machine Learning at Network Layer

Abstract

Abstract— Efficient packet transmission at the network layer is crucial for maintaining optimal performance in modern communication networks. Network congestion, caused by excessive data flow through intermediate nodes, often leads to packet delays, loss, and degraded throughput. This paper presents a Packet Transmission Analysis System that employs machine learning techniques to monitor and predict congestion at the network layer. A Python-based simulation module generates synthetic network traffic using varying topologies and protocols, including TCP and UDP. Packet- level parameters such as delay, loss ratio, throughput, and queue length are extracted and processed through a structured feature engineering pipeline. These features serve as input to multiple supervised learning models — Random Forest (RF), Support Vector Machine (SVM), and Long Short-Term Memory (LSTM) — trained to classify network states into Low, Medium, or High congestion levels. The trained models are integrated with a real-time Flask web interface, which displays live predictions, performance trends, and alerts. Experimental results demonstrate that the system can effectively analyze packet transmission behavior and accurately forecast congestion states. By combining simulation-driven data generation, feature extraction, and intelligent learning algorithms, the proposed framework enables proactive congestion detection and provides deeper insights into packet transmission dynamics at the network layer. Keywords: Network congestion prediction, Machine learning, Packet transmission analysis, Flask web interface.