The Highway Safety Manual (HSM), which is the guidance document for state departments of transportation (DOTs), was published in 2010, and one of its sections, called Part C of HSM, it involves the development of crash prediction models. However, HSM's default prediction models are not suitable for all states or jurisdictions because each state and jurisdiction have different characteristics, such as terrain, driver behaviors, weather conditions, etc. Hence, the principal objective of this study is to develop a prediction method for producing Ohio-specific safety performance functions (SPF) models to use for rural two-lane highways in the state of Ohio. Highway geometric data for almost 40,067 segments that have 21,666.03 miles and 79,481 total crashes that occurred for 4 consecutive years (2012-2015) were obtained from the Highway Safety Information System (HSIS) to create these new models using negative binomial regression and the pruned forward selection method by adding the interaction terms via JMP Pro software.

The most critical variables used for analyzing and creating the best models for the state of Ohio are average annual daily traffic (AADT), segment length, lane width, shoulder width, posted speed limit, presence of curves and grades, which were proven to be statistically significant in developing SPFs. Besides, the standard goodness-of-fit parameters were chosen to evaluate the regression models was AIC. Two models were created for rural two-lane road segments in the state of Ohio, which can be used to predict all crash types and fatal and injury crashes.