Answers are grounded in this paper and cite their sources.
In this paper, we propose a categorical model estimation framework with built-in feature selection, applied to traffic injury data from accident reports in France’s overseas territories. We address the problem of analyzing accident reports using categorical models, with the key challenge of identifying the most influential explanatory variables, thereby reducing model dimensionality. To overcome this challenge, we propose a methodology that integrates categorical model estimation with genetic algorithms (GA). The key elements of the proposed approach are: (i) estimating a categorical model on bootstrapped subsets of explanatory variable combinations; (ii) evaluating each subset’s model performance; (iii) selecting the top-performing subsets for GA-based optimization, using Goodman and Kruskal’s lambda as the optimality criterion, adjusted crossover strategies, and criterion stabilization as the termination condition; and (iv) predicting the target variable using the estimated optimal model configuration. Results show that our approach effectively isolates critical variables explaining traffic injury severity and provides a competitive, efficient, and interpretable statistical alternative to machine learning and deep learning models. The main findings of the presented framework relate to enhancing the automated assessment of traffic accident reports through a database-tailored categorical model. This offers a cost-effective solution that reduces manual evaluation efforts and improves consistency in injury severity classification.
No takes yet. Share an insight, caveat, or question.
Uglickich et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: