Predicting Open Parking Space using Deep Learning and Support Vector Regression

Vehicle parking issues have been one of the biggest problems faced in urban areas, as the supply and demand for vehicles and parking spaces are getting unbalanced year by year. The traditional approach of adding more parking spaces is no longer an effective solution. A practical and intelligent solution is to predict open parking spots using machine learning (ML), which would increase the utilization of available parking spaces and alleviate traffic congestion and decrease emissions from idling vehicles. This study aims to propose a parking prediction model using support vector regression (SVR) to predict available parking spaces. The data used in training the ML model is collected using a custom object detector, which is developed using the YOLOv4 (You Only Look Once) algorithm. The result shows that the custom YOLOv4 model is able to detect and identify empty and occupied parking spaces, and the SVR prediction model can predict the number of empty parking spaces. Two additional ML algorithms, which are linear regression (LR) and decision tree regressor were applied in this project to compare the performance of the SVR prediction model.