Analysis of different machine learning approaches in the context of urban mobility: a systematic review

In the urban planning of cities, with population growth and expansion of infrastructure demands leading to the displacement of people, the theme of urban mobility is highlighted in social agendas. As a result, in recent years, there has been an increase in research in the academic environment aimed at minimizing adversities, such as slow traffic flow, lack of accessibility in public and private equipment, reduction of long routes, and provision of means of transport. With the addition of large volumes of data, machine learning has contributed positively to solutions to combating these problems. From this perspective, this study sought to systematically review works in the academic literature that used machine learning in urban mobility, seeking to understand the main objects of study, data collection, learning techniques and algorithms, and evaluation metrics. With this work, after filtering and selecting studies, a continuous uniform exploration of the theme in the last four years was noticed. In addition, in each analyzed research, specific peculiarities were found in the intended objects of study. However, quantitatively, there was a greater preference for classification techniques, with the most commonly used algorithms being random forest, SVM, Decision Tree, convolutional neural network, KNN, and Naive Bayes. In this context, a small number of algorithms and metrics were identified in most studies, which may suggest a tendency.