This research project provides NCDOT with a planning-level safety analysis method and tool for estimating crashes at signalized intersections. The method uses Movement-based Safety Performance Functions (MBSPFs) that estimate crashes based on conflict point types and movement based volumes while also considering other traffic control and geometric features present at an intersection. A dataset of 282 signalized, 3 and 4-leg, intersections was utilized from Charlotte, NC. Several model forms were considered with the most promising forms including negative exponential, negative binomial, and Hurdle. Although model forms that used traffic control and geometric variables were more statistically significant, our research team found that simple MBSPFs based on traffic volumes were robust enough when compared to the more data-intensive models including intersection features. Cumulative residual (CURE) plots found showed that, when considering major and minor road AADT, the overwhelming majority of predictions were within the bounds of the 95% confidence interval – the only exceptions were for some volumes on the major road between 10.000 and 20,000 AADT. When testing Highway Safety Manual (HSM) predictions versus MBSPF models, HSM crash predictions were approximately double the root mean square error (RMSE) and mean absolute percentage error (MAPE), indicating that MBSPF was able to predict crashes more accurately. Last, an Excel-based tool was provided as part of the project to assist NCDOT with crash prediction for HSM and MBSPF models. The outputs from this tool related to MBSPFs allow for identification of high risk movements in addition to predicting overall crashes because the predictions are made at the individual conflicts.
