Conflict based crash prediction models for non lane based heterogeneous traffic at urban intersections in Dhaka city

Abstract

Road crash has become one of the major issues in Bangladesh recently. The problem is acute in urban areas especially in Dhaka metropolitan area (DMP). Among the road elements, intersections play a significant role in traffic crashes in this area. They contributed up to 40% of total crashes occurred in DMP area from 2013 to 2015. Historical crash data are used to assess road safety interventions of these intersections which often experience a lack of good and reliable data and longer waiting time. Different surrogate measures or indirect safety performance indicator can mitigate these drawbacks. An indirect safety performance indicator termed as Traffic Conflict Techniques (TCTs) are being used globally for lane based and homogenous traffic to predict the likelihood of crashes before they occur. In this study similar techniques have been applied for heterogeneous non lane based traffic situations. The research involves with 21 intersections in Dhaka city where 10 are 4-leg and 11 are 3-leg intersections. The TRRL of UK method was used to collect observed conflicts at selected intersections by trained observers. The conflicts were categorized according to seriousness of the evasive actions. Traffic volume of the study intersections was also collected during conflicts count. Based on ARRB’s technical manual the influence zones of the intersections were determined from the map and later the crash data of these influence zones were extracted from MAAP5 database of the Accident Research Institute (ARI). Analysis of conflict data shows that total and severe pedestrian conflicts were found to be highest (134 and 46.40 conflicts per hour respectively) at the 4-leg intersections. At 3-leg intersections, right turn same direction type conflicts for both categories were found to be dominating (159.27 and 40.00 conflicts per hour respectively). The crash data demonstrate that the 3-leg intersections yielded on average 3.02 crashes per year per intersection with standard deviation of 1.456 crashes per year, while the 4-leg intersections yielded average 1.25 crashes per year per intersection with standard deviation of 1.045 crashes per year per intersection. Crash to conflict ratios have been statistically determined for several types of collisions and conflicts for each of two types of intersections (4-leg and 3-leg). Total pedestrian crash-conflict ratios were highest than the others for both intersections and conflicts types. These crash-conflict ratios were used to develop the prediction models which had been presented through a number of equations. These ratios and equations can be applied to comparable intersections to obtain an expected crash rate of a specific type after the appropriate conflicts data are collected. The validity of the models was determined by random selection of the intersections and statistical correlation tests. The study ranked the risks of the selected intersections based on conflicts per unit volume data. It also established several correlations among crashes, volume and influence zones of the study intersections. Such techniques can be applied for identification of potential crash locations and hence design improvement, traffic management, and development of countermeasures at the intersections of Dhaka city.