Contextual Prediction of On-Street and Off-Street Parking Lot Utilisation: An Opening Gate Towards Sustainable Parking

There is a growing requirement for sustainable management of urban assets with the help of insights and trends into parking utilisation. Also, there is a need for faster discovery of available parking spaces through parking utilisation prediction, thereby reducing the environmental impact in cities. This paper presents models for on-street and off-street contextual parking utilisation prediction, using weather data and point-of-interest data as contextual data. CatBoost and Random Forest as machine learning (ML) methods are used for contextual parking utilisation prediction, and data analysis showed that models with contextual data achieved better results for both types of parking lots. In both cases, on-street and off-street parking, the best model is based on CatBoost and point-of-interest data, achieving R
²
92.84% in the case of on-street parking and 97.04% for off-street parking. Models with weather data performed lower than models with point-of-interests, where the difference between the models is 4.01% for on-street parking and 0.69% for off-street parking. The paper's overall conclusion is that ML models with contextual data are performing better than models without contextual data, and point-of-interest context provides better performance than weather data for both on-street and off-street parking.