Congestion or No Congestion: Packet Loss Identification and Prediction Using Machine Learning

• DOI: 10.48550/arxiv.2408.03007
• Publication Date: 2024-08-06
• Authors: Inayat Ali; Seungwoo Hong; Taesik Cheung
• Keywords: Identification; Packet loss

Packet losses in the network significantly impact performance. Most TCP variants reduce transmission rate when detecting packet losses, assuming congestion, resulting lower throughput and affecting bandwidth-intensive applications like immersive applications. However, not all are due to congestion; some occur wireless link issues, which we refer as non-congestive losses. In today's hybrid Internet, packets of a single flow may traverse wired segments reach their destination. should react same way it does congestive currently can differentiate between these types lowers its irrespective loss type, for clients. To address this challenge, use machine learning techniques distinguish at end hosts, utilizing easily available features host. Our results demonstrate that Random Forest K-Nearest Neighbor classifiers perform better predicting type loss, offering promising solution enhance