Application of clustering techniques in defining level of service criteria of urban streets

Abstract

The speed ranges for Level of Service (LOS) categories are not well defined for highly heterogeneous traffic flow on urban streets of India. The LOS analysis procedure followed in India is that developed by HCM 2000. The LOS categories for various urban street classes defined by HCM are apposite for developed countries having homogeneous type of traffic flow. For developing countries like India where the traffic flow is highly heterogeneous, LOS should be defined correctly taking into account the traffic and geometric characteristics. In this study an attempt has been made to define the LOS criteria of urban streets. Mumbai the business capital of India was chosen as the study area comprising of 100 street segments on four north-south and one east-west corridor. Second-wise speed data collected using Global Positioning System (GPS) receiver fitted on mobile vehicles was used for this study. Free-flow speed (FFS) data, average travel speeds during both peak and off peak hours and inventory details were collected and used in this study. These data are obtained from secondary source for this research work. Defining level of service is basically classification problems. Cluster analysis is found to be the most suitable technique for solving these classification problems. Four clustering methods namely Clustering Large Applications (CLARA), Self Organizing Tree Algorithm (SOTA), Hard Competitive Learning (hardcl) and Neural gas (ngas) were used to define LOS criteria in this study. Calinski-Harabasz Index, Homogenity Index, Stability Index, Connectivity Index, Average proportion of non-overlap Index, Average distance Index, Average distance between means Insex, Figure of merit Index, PtBiserial Index, Tau Index, GPlus Index, Ratkowsky Index, Duda Index, McClain Index are used in deriving optimum number of clusters.