Geostatistical approach to estimate car occupant fatalities in traffic accidents

Abstract

Despite the scientific and technological advances toward road crash prediction models, modelling road crashes in Brazil is a challenging task due to unreliable data and unavailability of essential information. Geostatistical approaches fit into this context as they not only incorporate the spatial factor, but also estimate variables in locations where they are not sampled. In order to contribute to this investigation and present geostatistics as a suitable tool in estimating road deaths, this study aimed to explore two different univariate interpolation methods to predict car occupant fatalities. In this study, we considered ordinary kriging and indicator kriging as such approaches. Analyses were held based on data from the state of São Paulo in Brazil. The results revealed a statistical outperformance in favor of indicator kriging, although spatial patterns found on kriging maps for both techniques indicated similarities in terms of hotspots. Furthermore, they were coherent with local aspects observed in the state, for instance those related to highway and vehicle characteristics.