Avaliação dos custos de manutenção de uma frota de veículos usando aprendizado de máquina e dados de telemetria das condições de condução

Abstract

This bachelor’s final project investigates the potential of vehicle fleet monitoring technologies to relate driving behavior with maintenance costs. Two databases are used: i) driving data obtained from fleet monitoring devices and ii) maintenance data provided by accredited service centers. Driving data characterize “offenses”as irregular driving patterns, such as abrupt braking and sharp turns. Maintenance data is derived from manually recorded service orders that often contain non-standardized informal terminology. In this case, natural language processing techniques are employed. Two regression models and one classification model were developed to relate maintenance costs to driving offenses and traveled distances. Regression models are intended to directly estimate maintenance costs, but their results are not promising. In contrast, the classification model showed superior performance by effectively categorizing maintenance costs as moderate or high, based on predefined criteria.