Bayesian crash prediction models to assess the safety risk of urban inntersections in metro Dhaka

Abstract

Bangladesh, having a serious and worsening road safety problem, is burdened with most of its urban crashes in the intersections. In order to have a better understanding of the factors affecting the probability of crashes in the intersections of Dhaka, the capital city, this research aims at developing several crash prediction models (CPMs) under empirical Bayesian (EB) framework. In addition, it also to provide directions for efficient use of resources applied to safety improvements by identifying and prioritizing crash prone intersections using EB technique. An extensive volume of dataset consisting the intersection crash data, intersection traffic exposure data, and intersection geometric and non-geometric data have been collected, synthesized, and investigated to determine the extent of contribution of each of the parameters to crashes and their statistical significance. A stepwise forward addition process was used to derive the covariates and finally, three CPMs were developed- (i) CPM for total crashes, (ii) CPM for severe crashes and (iii) CPM for minor and PDO crashes. In terms of goodness of fit measures, the scaled deviance and pearson chi-square have been found less than critical chi-suare value at 95% confidence interval for all the models while in terms of p-value of the parameters, CPMs for ‘total’ and ‘minor and PDO’ crashes have been found significant at 95% confidence interval while CPM for Severe crashes has been found significant at 90% confidence level. For all severity levels of crashes, positive correlations have been found with total entering traffic volume at the intersection and bus passengers’ activities on minor roads, while negative relationship with crashes have been found for the presence of on-street parking on major road. Total road crashes are likely to be increased by around 5.3% for 10% increase in traffic volume while presence of on-street parking on major road is causing a 34.2% reduction in predicted total collision. Low bus passenger activity has been found having relatively higher effect on crashes than the high bus passenger activity. Overall, modelling efforts for the crashes in the intersections identified several unique parameters particularly prevailing in low and middle income countries and also quantified their effects. Such findings have the potential to provide new insights in the occurrences of intersection crashes from the perspective of urban intersection crashes. EB technique being the current state-of-the-art practice, has been used in prioritizing the hazardous intersections explored in this research. Initially the relevant data were plugged into the CPMs to determine the predicted crash frequency (μi). Then with the help of appropriate mathematical procedures, expected crash frequency (EBi) for each of the intersections were determined. Finally, the intersections selected for the study have been ranked by sorting the expected crash frequencies in descending order. A relative comparison of prioritizing the crash prone intersections by different methods has also been demonstrated in this thesis. From model application perspective, the study recommends that the road safety authority uses the developed CPMs to assess the safety level of the urban intersections. Finally, this study also advocates that allocations of resources on road safety be based on the prioritized list of crash prone intersection by EB method instead of the traditional methods currently practiced in Bangladesh.