Urban traffic volume forecasting using machine learning and neural networks
TL;DR
The study found a positive relationship between traffic volume and temperature, while traffic volume decreased significantly on weekends, and the models with the highest prediction accuracy were XGBoost, Artificial Neural Network, and Ridge Regression.
Abstract
To build sustainable cities, it is critical to be able to predict traffic volume and to manage and control it effectively, especially in large cities. This study aims to determine whether the number of vehicles on the road in Istanbul, the world's most congested city, can be predicted using meteorological and traffic-related variables. The data used in this study were obtained from IBB Open Data Portal and the Weather Underground platforms. Machine learning techniques and Artificial Neural Networks were used in the research, and R programming language was used for data analysis. As a result of the research, the models with the highest prediction accuracy were XGBoost, Artificial Neural Network, and Ridge Regression, respectively. The most important variables affecting traffic volume are average speed, maximum speed, minimum speed, weekends, and temperature, respectively. The study also found a positive relationship between traffic volume and temperature, while traffic volume decreased significantly on weekends. The research results can guide vehicle drivers on what time of day and route to use and to pedestrians on whether they should use their personal vehicles or public transportation. On the other hand, local governments can manage increases in traffic volume at certain hours by evaluating whether the existing road infrastructure is sufficient for the observed daily traffic volume and peak-hour demand, thus contributing to sustainable urban life with less air pollution, less stress, fewer traffic violations, and fewer accidents.