Comparação de modelos de predição de categorias de acidentes nas rodovias federais

Abstract

Accidents on Brazilian federal highways are common and frequent, especially at certain times of the year and more specifically on certain highways. This work presents the application of Machine Learning techniques in public databases of the Federal Highway Police. The objective here is to predict future accidents based on models generated from Machine Learning. The original database was then enriched with data on tolls, speed cameras and national holidays. 32 tests were carried out with different combinations of algorithms and parameters. Although promising, the results show that accident prediction is a complex task and requires more data to achieve higher accuracy rates.