Extensive hypothesis testing for estimation of crash frequency models

• DOI: 10.1016/j.heliyon.2024.e26634
• Publication Date: 2024-02-23T19:22:34Z
• Authors: Zeke Ahern; Paul Corry; Wahi Rabbani; Alexander Paz
• Keywords: Overfitting; Benchmark (surveying); Identification

Estimating crash data count models poses a significant challenge which requires extensive knowledge, experience, and meticulous hypothesis testing to capture underlying trends. Simultaneous consideration of multiple modelling aspects is required including, among others, functional forms, likely contributing factors, unobserved heterogeneity. However, model development, frequently affected by time can easily overlook crucial such as identification necessary transformations, distributional assumptions. To facilitate development an estimation that extract many insights possible, optimization framework proposed generate simultaneously test diverse array hypothesis. The comprises mathematical programming formulation three alternative solution algorithms. objective function involves minimizing the Bayesian Information Criterion (BIC) avoid overfitting. algorithms include metaheuristics deal with NP-hard problem search through complex nonconvex space. also enable handle unique datasets varying strategies. effectiveness was ascertained using distinct datasets, published used benchmarks. results highlighted ability estimate models, surpassing benchmark in terms goodness-of-fit. provides several advantages, robust testing, uncovering specifications vital data, leveraging existing knowledge enhance efficiency. exposes vulnerability traditional analyst efforts fall into local optima, bias, limitations creating more efficient models. In compelling example from Washington, unveiled overlooked model, identifying speed, interchanges, grade breaks contributors, revealing potential danger excessively wide shoulders. Conversely, identified fewer factors missed non-linear relationship between safety shoulder widths. While wider shoulders are typically associated improved safety, suggest threshold beyond further widening could decrease safety. introduction random parameters analysis revealed nuanced frequency, thereby underlining incapable capturing