Building Cross-Validated Injury Risk Functions for Car-to-Car Impacts from Pre-Crash Predictors Only

Summarise with VIRAT
Authors: Giulio Vichi, Michelangelo-Santo Gulino, Dario Vangi
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

ABSTRACT: In car-to-car collisions, injury risk models are widely used to inform road-safety decisions because they complement analyses of crash frequency and nominal severity with probabilistic estimates of harm to occupants. However, most established injury risk functions (IRFs) in the literature rely on crash-phase variables (e.g., metrics derived from impact dynamics as ∆V or EES) and therefore require computationally intensive simulation or detailed retrospective crash reconstruction. This limits their applicability in time-critical settings and accident analysis where only pre-crash information is available. This study proposes and evaluates maximum abbreviated injury scale (MAIS)-related IRFs based solely on pre-crash predictors to estimate risk to occupants of a passenger car struck by another car. The most informative pre-crash variables (e.g. closing speed or kerb weight) are identified using feature ranking and greedy sequential selection within a systematic pipeline. Generalizability is assessed through cross-validation across three geographically distinct in-depth accident databases: the international IGLAD dataset, the U.S. NHTSA/NASS CDS, and the French VOIESUR dataset. Applied to these datasets, the resulting IRFs for severe and fatal injuries (MAIS 3+) exhibit strong predictive accuracy and deliver reliable estimates across regions. By contrast, IRFs targeting any injury (MAIS 1+) do not show robust classification performance. These findings indicate that pre-crash information suffices for dependable estimation of severe and fatal outcomes, whereas probabilistic discrimination of minor injuries may require crash-phase features or additional context. Because the proposed IRFs rely solely on pre-crash variables and show geographical robustness, they are well suited to time-critical applications —such as supporting emergency decision-making in automated driving— and strategic safety analyses where crash reconstruction is not feasible.

Fields of Study & Subjects

Traffic and Road Safety Automotive and Human Injury Biomechanics Transportation Safety and Impact Analysis

Publication details

Journal / source
Florence Research (University of Florence)
Publication date
1 January 2026
Language
EN