Análise de dados para identificação de atividades de estacionamento de veículos por meio de aprendizado de máquina

Abstract

The difficulty of remembering where a vehicle was parked is a common problem, especially in unfamiliar environments. This study aimed to develop an analysis and investigation regarding the possibility of distinguishing parking activities and vehicle motion through machine learning algorithms. A mobile application was developed using the Flutter framework to collect data through sensors such as the speedometer, gyroscope, and magnetometer, in order to train machine learning algorithms. The k-NN, Random Forest, and SVM algorithms were employed to analyze the collected data. The research was conducted in two phases, using distinct datasets for training and testing. The performance of the algorithms was evaluated based on the accuracy achieved in classification. In summary, the results revealed that the Random Forest algorithm achieved the highest efficacy in classifying vehicle parking, reaching an accuracy of 92.71% in the first phase and 94.12% in the second phase. These results highlight the capability of the proposed system to accurately distinguish parking activities in different contexts.