An optimization framework for crash count data models

This study presents an optimization framework to address the complexities involved in estimating crash frequency models. The purpose is to efficiently generate and examine a diverse range of model specifications to capture underlying patterns and likely contributing factors. The framework incorporates a mathematical programming formulation and a metaheuristic approach to minimize the Bayesian Information Criterion (BIC) and identify potential model configurations, aiming to provide a deeper comprehension of the data and overcome the limitations of conventional model development approaches. The proposed framework offers to enhance the estimation of crash count data models and provides numerous benefits, including extensive hypothesis testing and uncovering significant insights that have the potential to be disregarded due to restricted or biased hypothesis testing.