From contextual to global rankings by passive safety of generational classes of light vehicles

Zaïd Ouni (1) , Cyril Chauvel (2) , Antoine Chambaz (1)

1 MODAL'X - Modélisation aléatoire de Paris X
2 Laboratoire d'Accidentologie et de Biomécanique

Each year, the BAAC (Bulletin d'Analyse des Accidents Corporels) data set includes traffic accidents on French public roads involving one or two light vehicles and injuring at least one of the passengers. Each light vehicle is associated with its "generational class" (GC), which gives a raw description of the vehicle. Two light vehicles with two different GCs do not necessarily offer the same level of passive safety to their passengers in different contexts of traffic accident. The objective of this study is to assess to which extent more recent generations of light vehicles are safer than older ones based on the BAAC data set. In an ealier study, we elaborated an algorithm for the contextual ranking of GCs. In the present study, our objective is to develop an algorithm for the global (as opposed to contextual) ranking of GCs. Like in the earlier study, we rely on "scoring": we look for a score function that associates any GC with a real number; the smaller is this number, the safer is the GC across all contexts of accident. Causal arguments help to formalize our objective in statistical terms. We rely on cross-validation to select the best score function among a collection of candidate score functions which are built based on the algorithm for the contextual ranking of GCs and a collection of working models. We implement the resulting algorithm, apply it, and show some results.

Mots clés: car safety; causal analysis; cross-validation; scoring

Domaines: Statistiques [math.ST]

Zaïd Ouni, Cyril Chauvel, Antoine Chambaz. From contextual to global rankings by passive safety of generational classes of light vehicles. 2016. ⟨hal-01359225⟩