Creative Commons

With rapid population growth and development, traffic congestion has become a prob-lem over the past decades, especially in the big cities. Traffic speed is one of the indica-tors to determine traffic conditions. With the current development of sensors and other data collection devices and the advancement of the Internet of Things (IoT), more traf-fic speed data has become available. At the same time, there is growing interest in Intel-ligent Transportation Systems (ITS) that may help solve the current traffic problem. Traffic speed prediction is one of the components in ITS that has become one of the most developed research areas. However, most of the studies used Deep Learning (DL) algorithms which require complex data preprocessing and extensive computational pow-er. Moreover, many studies did not consider the spatial features and only predicted the speed in a road corridor or the segment where a sensor is available. Therefore, this study is trying to compare the Machine Learning (ML) performance to an earlier study by con-sidering the spatial features and finding the most effective way to use these models for network-wide prediction. Floating Car Data (FCD) and OpenStreetMap (OSM) data were used and prepro-cessed by following the approach of the previous study. The data were trained and test-ed using several ML models. The result was evaluated by comparing it to the initial study's results. Moreover, several scenarios were tested to determine the most efficient way to predict the traffic speed on a network-wide scale. The Gradient Boosting (GB) algorithm is the best model in this study, with consistent performance in all scenarios and promising results.