| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 4.14 MB | Adobe PDF |
Advisor(s)
Abstract(s)
A Polícia Rodoviária Federal (PRF) é o órgão responsável pela fiscalização de trânsito no âmbito das rodovias federais brasileiras. Uma das suas missões é garantir a segurança viária e proteger a vida dos usuários das rodovias. Um dos seus grandes desafios é identificar causas sistemáticas de acidentes, sejam estruturais ou sociais, e atuar para reduzir o número de ocorrências, por meio de políticas e patrulhamento ostensivo. O objetivo desse trabalho é realizar um estudo da utilização de técnicas de Machine Learning (ML) para a identificação dos locais com maior probabilidade de ocorrência de Acidentes de Trânsito (AT), a partir dos dados históricos de acidentes da PRF. Para isso, é analisada a capacidade de diversos algoritmos de ML em predizer o local de ocorrência de um novo AT. Além disso, é verificado se a utilização de outras variáveis que não são consideradas no método atual da instituição, tais como a condição climática, o tipo de acidente e a geolocalização do local da ocorrência, influenciam na assertividade dos algoritmos.
The Polícia Rodoviária Federal (PRF – Federal Highway Police) is responsible for supervising traffic on Brazilian federal highways. One of its missions is to ensure safety and protect the lives of road users. One of its great challenges is to identify systematic causes of accidents, whether structural or social, and to act to reduce the number of occurrences, through policies and ostensible patrolling. The objective of this work is to carry out a study of the use of Machine Learning (ML) techniques to identify the places with the highest probability of occurrence of Traffic Accidents (TA), from the historical accidents data of PRF. For this, the ability of several ML algorithms to predict the place of occurrence of a new TA is analyzed. In addition, it is verified whether the use of other variables that are not considered in the institution's current method, such as the weather condition, the type of accident and the geolocation of the place of occurrence, influence the assertiveness of the algorithms.
The Polícia Rodoviária Federal (PRF – Federal Highway Police) is responsible for supervising traffic on Brazilian federal highways. One of its missions is to ensure safety and protect the lives of road users. One of its great challenges is to identify systematic causes of accidents, whether structural or social, and to act to reduce the number of occurrences, through policies and ostensible patrolling. The objective of this work is to carry out a study of the use of Machine Learning (ML) techniques to identify the places with the highest probability of occurrence of Traffic Accidents (TA), from the historical accidents data of PRF. For this, the ability of several ML algorithms to predict the place of occurrence of a new TA is analyzed. In addition, it is verified whether the use of other variables that are not considered in the institution's current method, such as the weather condition, the type of accident and the geolocation of the place of occurrence, influence the assertiveness of the algorithms.
Description
Keywords
Acidente de trânsito Acidente rodoviário Rodovias federais Inteligência Artificial Aprendizagem Automática Aprendizado de máquina Polícia rodoviária federal PRF Traffic accident Road accident Highway accident Federal roads Brazilian highway Federal highway Artificial intelligence Machine Learning Federal highway police Deep learning