Applications of Machine Learning Methods in Macroscopic Crash Analysis for Transportation Safety Management

Transportation Safety Planning (TSP) is a statewide-scale tool and combines transportation planning processes with safety aims to increase safety and reduce transportation fatalities and injuries. Traffic safety, which continues to remain a critical issue worldwide, has led to a myriad of modeling techniques to improve analytical capabilities with respect to crash modeling and prediction. State and metropolitan transportation planning processes must be consistent with Strategic Highway Safety Plans. This research aims to identify models and methods to improve the ability to capture variables that have the most significant impact on traffic safety through crash prediction modeling. In order to achieve this research goal, the research objectives are as follows: Identify important variables in TSP. Investigate different areal unit such as traffic analysis zones (TAZs) and traffic analysis districts (TADs). Explore the modifiable areal unit problem (MAUP), which addresses crashes on the boundaries and autocorrelation in macro-level crash modeling. Analysis of before and after crashes and testing Poisson distribution This research explores the application of parametric and nonparametric approaches to use different models for prediction and inference, with the aim of minimizing the reducible error. Since a macro-level analysis involves aggregating crashes per spatial unit, a spatial dependence or autocorrelation may arise if a variable of a geographic region is affected by the same variable of the neighboring regions. So, this study also will explore the effect of spatial autocorrelation in modeling crashes in TAZs and TADs