Predicting open space parking vacancies using machine learning

Lee, Wei Jun (2023) Predicting open space parking vacancies using machine learning. Final Year Project, UTAR.

Preview

PDF Download (4Mb) | Preview

Abstract

Vehicle parking has become a significant issue in urban areas due to the imbalance between supply and demand for parking spaces, and increasing the number of parking spaces is no longer an effective solution. Predicting open parking vacancies using machine learning is a practical and effective solution to overcome parking issues. The ability to predict parking availability maximizes parking space utilization, ultimately alleviating traffic congestion. The reduction in idling vehicles results in a decrease in gas emissions, which reduces the burden on the environment. This study proposes a parking prediction model using support vector regression (SVR) to predict available parking spaces. A custom object detector developed using the YOLOv4 algorithm was used to collect the data for training the machine learning model. The results show that the custom YOLOv4 model accurately detects and identifies empty and occupied parking spaces, while the SVR prediction model can predict the number of empty parking spaces. Noise such as weather, lightning issue and obstacles is considered in YOLOv4 model. Next weather features is included in training the machine learning model. In this project, two additional machine learning algorithms, namely linear regression (LR) and decision tree regressor, were used to compare the performance of the support vector regression (SVR) prediction model. Additionally, four different hyperparameter tuning techniques were employed to obtain the most promising fine-tuned support vector regression (SVR) model, including grid search, random search, random search plus, and parameter optimization loop. Moreover, a PySimpleGUI was developed to provide an interactive parking vacancy prediction model graphic user interface (GUI).

Actions (login required)