Driver Pattern Identification in Road Crashes in Spain

Abstract

The authors thank the Community of Madrid for its partial funding which, through the SEGVAUTO 4.0-CM (P2018/EMT-4362) program, has contributed to its development and dissemination.

Road Safety as a Public Health Problem: Case of Ecuador in the Period 2000–2019

“…The chi-square analysis of the relationship between the variables of sex of the victims involved in traffic accidents and injury severity shows that the variables are independent, as concluded by other studies [23][24][25][26][27]59]. A recent work concerning gender in road safety has applied clustering analysis to the pattern identification in road crashes with two passenger cars in Spain [39]. The comparison of traffic crashes between the morning (0:00 to 11:59 h) and the afternoon (12:00 to 23:59 h) showed that the decrease in traffic crashes between the two schedules was parallel (p = 0.10) with a total reduction of −8.63 per year as shown in Figure 2.…”

“…However, exposure data are hard to collect and unreliable, and recourse to alternative methods such as indirect or quasiinduced exposure is necessary [25,32,[35][36][37][38]. The investigation [39] driver collision patterns addressed by sex and age regarding offenses, collision type, and injury severity using clustering methodology and tests of proportions are used for a complimentary analysis when driver offense is present, concluding there are significant differences between males and females, showing that this tool is very useful to support conclusions.…”

Driver Liability Assessment in Vehicle Collisions in Spain

“…Here, the choice of 25 nodes was made empirically by trial and error: SOMs with different map sizes have been obtained, starting with the smallest ones. 25 nodes were considered a reasonable choice, given the trade-off sought between properly identifying patterns (clarity/visibility) and sample size per cluster: with an excessively small map size, clusters could be too heterogeneous and, therefore, adequate patterns would not be extracted; the same would occur with a very large map size, resulting in too small cluster sample sizes [39]. Also, 25 clusters is an adequate number to provide a better understanding of the multivariate structure of the data.…”

“…Each circular sector radius within a cluster will be smaller or larger (the angle is the same for all variables, 360/number of variables) depending on the average value (over all the drivers in the cluster) of the variable it represents. The radius will be maximum when either (a) the value of the variable in question for all drivers is 2, which means that all the drivers in the cluster have this offence or impairment or (b) the average of the variable in this cluster is larger than any other one (cluster), whereas it will be minimal when the average is 0 (the circular sector is not represented for that variable) and, therefore, no driver in the cluster will have committed that offence or present the impairment that the variable indicates [39].…”

“…Therefore, these variables should perhaps not be taken into account in the liability assignment process. However, disability is mainly related with the age of the drivers [39] and elderly drivers are more likely than younger ones to be at-fault in a crash [15]. Thus it could be useful to include this variable in order to obtain more information about driver behavior patterns.…”