Performance Assessment and Modeling of Routing Protocol in Vehicular Ad Hoc Networks Using Statistical Design of Experiments Methodology: A Comprehensive Study

Abstract

The performance assessment of routing protocols in vehicular ad hoc networks (VANETs) plays a critical role in testing the efficiency of the routing algorithms before deployment in real conditions. This research introduces the statistical design of experiments (DOE) methodology as an innovative alternative to the one factor at a time (OFAT) approach for the assessment and the modeling of VANET routing protocol performance. In this paper, three design of experiments methods are applied, namely the two-level full factorial method, the Plackett–Burman method and the Taguchi method, and their outcomes are comprehensively compared. The present work considers a case study involving four factors namely: node density, number of connections, black hole and worm hole attacks. Their effects on four measured outputs called responses are simultaneously evaluated: throughput, packet loss ratio, average end-to-end delay and routing overhead of the AODV routing protocol. Further, regression models using the least squares method are generated. First, we compare the main effects of factors resulted from the three DOE methods. Second, we perform analysis of variance (ANOVA) to explore the statistical significance and compare the percentage contributions of each factor. Third, the goodness of fit of regression models is assessed using the adjusted R-squared measure and the fitting plots of measured versus predicted responses. VANET simulations are implemented using the network simulator (NS-3) and the simulator of urban mobility (SUMO). The findings reveal that the design of experiments methodology offers powerful mathematical, graphical and statistical techniques for analyzing and modeling the performance of VANET routing protocols with high accuracy and low costs. The three methods give equivalent results in terms of the main effect and ANOVA analysis. Nonetheless, the Taguchi models show higher predictive accuracy.