Application of genetic programming clustering in defining LOS criteria of urban street in Indian context

Abstract

India is a highly populated country having second largest road network in the world. Owing to boastfully population, the congestion is growing rapidly on the urban road networks. The level of service (LOS) is not substantially defined for heterogeneous traffic flow with different operational characteristics. Defining LOS is essentially a classification problem. The application of cluster analysis is the worthiest proficiency to solve such problem for which genetic programming (GP) clustering, an evolutionary algorithm is used in this study. Five cluster validation parameters are utilized to examine the optimal number of clusters. The cluster validation parameters are used to obtain the number of categories of urban street classes. After acquiring optimal number of clusters, GP clustering is implemented to the free flow speed (FFS) data to get ranges of different urban street classes. Again, GP clustering is enforced on average travel speeds of street segments to specify the ranges of different LOS categories. Speed data used in this study are collected using Trimble GeoXT GPS receivers fitted on mid-sized vehicles for five major urban corridors comprising of 100 street segments of Greater Mumbai region. Result shows that FFS of urban street classes and average travel speed of LOS categories are lower than that mentioned in Highway Capacity Manual (HCM 2000) on account of physical and surrounding environmental characteristics. Also, average travel speed of LOS categories expressed in terms percentage of FFS of urban street classes found to be different from that mentioned in HCM 2010.