Regras de associação entre as características dos veículos e os acidentes de trânsito em rodovias federais brasileiras através de aprendizado de máquina

• Ano de defesa: 2022
• Autor(a) principal: Ramon Batista de Araújo
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Minas Gerais; Brasil; ENG - DEPARTAMENTO DE ENGENHARIA TRANSPORTES E GEOTECNIA; Programa de Pós-Graduação em Geotecnia e Transportes; UFMG
• Palavras-chave em Português: Acidentes; Regras de associação; Aprendizado de máquina; Algoritmos; Transportes; Aprendizado de computador; Acidentes de trânsito
• Link de acesso: http://hdl.handle.net/1843/47414

Traffic accidents are pressing public health problems, which lead to a series of deaths and injuries, representing not only numbers, but lost human lives. In view of this, the social impact added to the costs with a significant number of deaths and injuries highlight the need for a deeper analysis of the causes of accident. For this reason, this research aimed at identifying association rules between the causes of accidents and the characteristics of vehicles, roads, users, and the environment on Brazilian federal highways, comparing the Apriori, Eclat, FPGrowth, and FP-Max machine learning techniques in data processing. To achieve this objective, the methodology of this research basically applied the use of categorical variables data tabulation, using a mixed method for data collection and transformation and analysis of results, through a procedure within a real and local context in a case study. In this way, through the analysis of the results, it was possible to compare the algorithms and, thus, verify that the Apriori, FP-Growth, and Eclat algorithms perform equally, with similar support and number of characteristics indexes, where the higher the number of characteristics, the lower the support index. On the other hand, the FP-Max, which proposes a greater support metric for a higher number of characteristics, achieved the opposite outcome, consequently providing a more accurate result. However, FP-Max, as well as Eclat, did not present lift and confidence indexes for the analyzed database. Therefore, taking these factors into consideration, it is possible to affirm that the collaboration of a method capable of understanding the causes of accidents can help public policies and be a strong social and scientific contribution. This research has a promising potential to be used as a basis for several studies, by the Federal Highway Police itself, safety engineers, public authorities, and also by the private sector such as highway concessionaires and mobile application developers.