On-street car parking prediction in smart city: a multi-source data analysis in sensor-cloud environment

Smart car parking systems in smart cities aim to provide high-quality services to their users. The key to success for smart car parking systems is the ability to predict available car parking lots throughout the city at different times. Drivers can then select a suitable car parking location. However, the prediction process can be affected by many different factors in smart cities such as people mobility and car traffic. This study investigates the use of multi-source data (car parking data, pedestrian data, car traffic data) to predict available car parking in fifteen minute intervals. It explores the relationship between pedestrian volume and demand for car parking in specific areas. This data is then used to predict conditions on holidays and during special events, when the number of pedestrians dramatically increases. A Gradient Boosting Regression Trees (GBRT) is used for prediction. It is an ensemble method that can be more accurate than a single Regression Tree and Support Vector Regression. The probability of error for our model is 0.0291.