Estimativas do nível de severidade dos acidentes envolvendo motociclistas em vias urbanas: uma análise das variáveis explicativas com redes neurais artificiais

Abstract

Traffic accidents on urban roads are considered as phenomena of great variability and, have drawn attention of researchers in the field of road safety in an attempt,both to understand what are the variables that contribute to its occurrence, and to estimate the likelihood of these events. Normally are used regression models, generally not linear; known as Accident Prediction Models (MPA), which tend to correlate the accident with the attributes of the road.The main objective of this work is to build a model based on artificial intelligence dedicated to the estimation of the variables that contribute most in the occurrence of different levels of severity of traffic accidents with motorcycles, the arterial roads in the city of Fortaleza. The model was calibrated, validated and tested using a sample of 1137 observations and explanatory variables; geometric attributes were used, climatic, operational and functional attributes of the route. Although the MPA are performing well in their predictions, the model implemented here, based on Artificial Neural Networks (ANN) gave considerably promising results.