Análise e modelagem dos acidentes de trânsitos nas rodovias federais no estado de Goiás entre 2017 e 2018

Abstract

Brazil is ranked among the countries with the highest number of deaths from traffic accidents in the world. In the Goiás State, traffic accidents that occur on federal highways are no exception to this statistic, as it has an intense vehicle flow and serves as a connection for people and commodities because it is located in the central region of Brazil. Therefore, it is important to identify the areas where these accidents occur most and to understand their possible triggers. This work aimed to analyze traffic accidents on federal highways that cross the Goiás State (Brazil) between the years 2017 and 2018 and, furthermore, to evaluate machine learning algorithms, in particular, based on decision tree in the prediction of these accidents. An exploratory analysis of the accident database from the Federal Highway Police was performed, focusing on the causes, types and classification of these accidents. Regarding prediction, the algorithms CART, C4.5, C5.0 and Random Forest were tested for the prediction of accidents with causes associated with the road conditions. The highest accident density occurred in the urban perimeter, especially in the routes between Goiânia City and the Federal District. The main causes of accidents are associated with human error, predominantly the types of leaving the carriageway, rearend collision, and rollover. For the accidents prediction related to road conditions, the best performance was achieved with the C5.0 and Random Forest algorithms. Variables associated with weather conditions and road surface infrastructure were the most important variables in predicting the occurrence of these accidents.