Uso de técnicas de aprendizado de máquina para identificação de perfis de uso de automóveis baseado em dados automotivos

Abstract

Traffic violence has caused great damage. And above all, victimized many citizens, whether drivers or non drivers. According to studies from brazilian National Road Safety Observatory (ONSV, 2017), 90% of traffic accidents are caused by drivers’ recklessness, 5% by car defects and 5% by poor road conditions. One of the alternatives to support actions that address these problems is to understand how car drivers behave when they are behind the wheel. Using vehicle information to understand drivers is an issue that has gained importance in recent years and, in the face of the problems involved, identifying car use profiles has increasingly been a subject of worldwide research. This work presents a model for the use of Machine Learning techniques (descriptive and predictive) on vehicle data obtained from OBD-II (On Board Diagnostics II) to identify possible profiles of automotive using. After the entire refinement process, the model presented more than 99% accuracy in the identification of 3 profiles (low, mid and high). To implement the model, a platform based on a distributed architecture (Web Server, Mobile Application and Service API) was created. This platform is able to capture data from a car and return its usage profile.