Machine Learning Models in the large-scale prediction of parking space availability for sustainable cities

Abstract

The search for effective solutions to address traffic congestion presents a significant challenge for large urban cities. Analysis of urban traffic congestion has revealed that more than 70% of it can be attributed to prolonged searches for parking spaces. Consequently, accurate prediction of parking space availability in advance can play a vital role in assisting drivers to find vacant parking spaces quickly. Such solutions hold the potential to reduce traffic congestion and mitigate its detrimental impacts on the environment, economy, and public health. Machine learning algorithms have emerged as promising approaches for predicting parking space availability. However, comparative studies on those machine learning models to evaluate the best suited for a large-scale prediction and within a given prediction time period are missing.In this study, we compared nine machine learning algorithms to assess their efficiency in predicting long-term, large-scale parking space availability. Our comparison was based on two approaches: using on-street parking data alone and 2) incorporating data from external sources (such as weather data). We used automatic machine learning models to compare the performance of different algorithms according to the prediction efficiency and execution time. Our results indicated that the automated machine learning models implemented were well fitted to our data. Notably, the Extra Tree and Random Forest algorithms demonstrated the highest efficiency among the models tested. Moreover, we observed that the Random Forest algorithm exhibited less computational demand than the Extra Tree algorithm, making it particularly advantageous in terms of execution time. Therefore, this work suggests that the Random Forest algorithm is the most suitable machine learning model in terms of efficiency and execution time for accurately predicting large-scale, long-term parking space availability.