Regional Map-Based Analytical Platform for Statewide Highway Safety Performance Assessment
Islam, SMA Bin Al
MetadataShow full item record
This research extended models for predicting current crash counts by severity (CCS) by developing a two-stage regression model and a generalized nonlinear regression model for formulating a new method for identifying CCS-based hotspots. The hotspot identification (HSID) method is designed to improve the safety performance module of the Digital Roadway Interactive Visualization and Evaluation Network (DRIVE Net) and to enable a regional, map-based, real-time analytical platform for statewide highway safety performance assessment. The most important contributing factors (static and dynamic) to traffic crashes of different severity types, including traffic characteristics, road conditions, and weather conditions, were identified by using the structured framework developed in this research. A total of 802 road segments on I-5, I-90, I-82, I-182, I-205, I-405 and I-705 in Washington state were selected as the candidate sites for data collection. A two-stage regression and logistics model was developed to predict crash counts on freeway segments by severity. The regression analysis found that annual average daily traffic per lane, number of lanes, curvature of segments, width of the outer shoulder, width of the inner shoulder, width of medians, average speed limit, lane surface type, outer shoulder type, inner shoulder type, and road surface conditions show strong relationships with the crash frequencies of different severity levels. A CCS-based HSID method was developed by employing a two-stage regression approach. A new safety performance index (SPI) and a new potential safety improvement index (PSII) were developed by introducing the risk weight factor and were compared with three indices by employing HSID evaluating methods. The results of four consistency tests revealed that the SPI method is the most consistent and reliable method for identifying hotspots. Finally, a generalized, nonlinear, model-based multinomial logistic regression approach was also developed to estimate the probability and frequency of crashes for different severity levels. It also showed that the significance and nonlinearity for each crash severity level are different among the contributing factors. This evaluation suggested that the SPI method (among the methods compared) has the potential to become the industry standard. Finally, a regional, map-based analytical platform was developed within the DRIVE Net system by expanding the safety performance module with the new SPI and PSII functions.