Hybrid dimension reduction and logit models for glare-induced crash severity

Tusti, Anannya Ghosh; Starewich, Michael; Barua, Swastika; Chowdhury, Tausif Islam; Polock, Sazzad Bin Bashar; Alnawmasi, Nawaf; Das, Subasish · 2026 · DOAJ

DOI: 10.1038/s41598-026-42745-z

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the safety risks associated with sun and headlight glare, which reduce contrast sensitivity and hazard detection, leading to severe crashes. Motivated by the lack of systematic classification of glare-related collisions and the need to account for unobserved heterogeneity in crash severity, the authors develop a two-stage, cluster-driven discrete choice framework. The research aims to identify distinct crash typologies and quantify how specific factors influence injury severity across these groups, thereby informing targeted countermeasures. The methodology utilizes Texas police-reported crash data from 2017 to 2024, comprising 11,326 crashes attributed to vision obstruction by glare. The first stage employs Cluster Correspondence Analysis (CCA) to segment the data into homogenous crash typologies, using variable importance rankings from XGBoost and Random Forest models alongside Cramér’s V correlation analysis for feature selection. The second stage estimates cluster-specific multinomial logit (MNL), random-parameter logit (RPL), and RPL with heterogeneity in means (RPLHM) models to assess injury severity outcomes: property damage only (PDO), non-incapacitating/minor injury (BC), and fatal/incapacitating injury (KA). Model fit was evaluated using likelihood ratio tests, AIC, and McFadden’s Pseudo R-squared, with simulated maximum likelihood estimation using 1,000 Halton draws. The results reveal pronounced cluster-dependent severity mechanisms. Three distinct crash typologies were identified. Low-speed urban angle crashes, particularly under daylight conditions, exhibited substantially lower risks of fatal and incapacitating injuries. Conversely, high-speed straight-line and rear-end crashes, especially those involving speed-control failures, showed the highest likelihood of severe outcomes. Rural nighttime crashes on unlit two-lane roads demonstrated elevated injury severity due to headlight dazzle, with vehicle type and lighting conditions moderating these effects. The RPLHM specification provided superior model fit compared to standard MNL, confirming the presence of latent heterogeneity. Key findings indicate that lower speed limits and daylight consistently reduce fatal injury risk, while higher speeds, inadequate lighting, and straight-line movements increase severity. The significance of this study lies in its demonstration that integrating clustering with random-parameter logit modeling improves interpretability and supports the design of targeted countermeasures. The findings suggest that generic safety strategies are insufficient; instead, interventions should be tailored to specific glare contexts. Recommended actions include dynamic speed management, glare-reducing lighting treatments, adaptive headlamp technologies, and focused driver education. By identifying distinct risk profiles, the study provides empirical evidence to inform headlight alignment policies, roadway design guidelines, and lighting standards, ultimately aiming to mitigate the disproportionate impact of glare on vulnerable populations and high-risk driving scenarios.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• roadway lighting effects
• disability glare
• weather rain fog snow
• visibility analysis litigation
• induced exposure
• discomfort glare

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Empirical Findings: crash risk outcomes