An Effective Framework for Packet Behavior Classification in High- Speed Connectionless Networks

Abstract

The surge in connectionless data transmissions, driven by IoT devices and streaming services, has intensified challenges such
as packet loss, jitter, and delay. Conventional monitoring systems often fail to adapt to these dynamic conditions, highlighting the need
for intelligent Machine Learning (ML)-based approaches. This paper presents a comprehensive ML framework for evaluating and
predicting packet behavior in connectionless networks using synthetically generated datasets. Two models—Random Forest and Support
Vector Machine (SVM)—were developed and compared. The Random Forest classifier demonstrated superior performance, achieving
98.33% precision, recall, and F1-score across all classes, while the SVM model yielded a slightly lower overall accuracy of 97%. Feature
importance analysis revealed packet loss rate and jitter as critical predictors influencing classification. Performance validation through
confusion matrices, ROC-AUC scores, and precision-recall curves confirmed the high reliability of both models, with each attaining an
AUC of 1.00. The findings establish Random Forest as the more robust and accurate choice for anomaly detection in unstable network
environments. Leveraging such ML techniques enables proactive monitoring, prediction, and optimization of network performance,
ultimately enhancing the reliability and efficiency of connectionless communication systems.