Machine Learning for Predicting Pedestrian Activity Levels in Cities

Abstract

Analysing and modelling pedestrian activity in built environments allows us to understand, assess, predict, and manage its dynamics. Nonetheless, pedestrian activity data might not be available everywhere. An alternative can suggest predicting pedestrian activity by considering environmental characteristics and the geometrical configuration of the environment. This paper presents a Machine Learning pedestrian activity level prediction model, which is trained and tested using data extracted from smart city sensor systems from multiple cities. The proposed model was applied to Greater London, UK, and the prediction results were compared with pedestrian activity data provided by Transport for London. Our results show that the model has high potential to predict pedestrian activity levels in a city, but that further research is needed to obtain more reliable results.