Random Forest-Driven Feature Importance Assessment for QoS in MPLS and SD-WAN

Abstract

This study presents a novel comparative analysis employing Random Forest regression to quantify the relative importance of key Quality of Service (QoS) parameters—packet loss, delay, and jitter—in Multiprotocol Label Switching (MPLS) and Software-Defined Wide Area Network (SD-WAN) architectures. Using empirical data collected from controlled simulations of multimedia traffic, the feature importance scores reveal that packet loss overwhelmingly dominates as the critical factor influencing network performance, with scores of 0.8620 in SD-WAN and 0.7259 in MPLS, indicating an 18.76% increase in SD-WAN’s sensitivity to packet loss. Delay exhibits moderate relevance in MPLS, with an importance score of 0.2205, but shows markedly reduced significance in SD-WAN at 0.1341 (a 39.21% decrease). At the same time, jitter demonstrates negligible influence across both networks, with scores below 0.054. These findings confirm that SD-WAN’s dynamic path optimisation effectively mitigates delay effects, whereas packet loss remains the principal constraint on performance. This work constitutes the first methodical Random Forest-based comparative evaluation of QoS parameter importance across MPLS and SD-WAN, delivering robust, data-driven insights tailored to each architecture’s operational characteristics. The framework provides network operators with critical guidance for targeted QoS optimisation, prioritising packet loss mitigation strategies, particularly within SD-WAN environments. Overall, this research establishes an empirical foundation for architecture-specific QoS management, advancing intelligent network performance assessment through machine learning techniques.