Abstract :
A substantial amount of research has been done to develop improved Intelligent Transportation Systems (ITS) to alleviate traffic congestion problems. These include methods that incorporate the indirect impact on traffic flow such as weather. In this paper, we studied the impact of weather conditions on traffic congestion along with more spatial and temporal factors, such as weekdays/time and location, which is a different approach to this problem. The proposed solution uses all these indicators to estimate the flow of traffic. We evaluate the level of congestion (LOC) based on the traffic volume grouped in certain regions of the city. The index for the defined LOC indicates the traffic flow from “free -flowing” to “traffic jam”. The data for the traffic volume count is collected from the Department of Transportation (DOT) for NYMTC. Weather conditions along with special and temporal information have an essential role in predicting the congestion level. We used supervised machine learning for this purpose. The prediction models are based on certain factors such as the volume count of the traffic at the entry and exit point of each street pair, particular days of the week, timestamp, geographical location, and weather parameters. The study is done on the major roadways of each of the four prominent boroughs in New York. The results of the traffic prediction model were established by using the Gradient Boosting Regression Tree (GBRT) which showed an accuracy of 97.12%. Moreover, the calculation speed was relatively fast, and it has stronger applicability to the prediction of congestion conditions.
Keyword :
Gradient Boosting; Decision Tree Algorithm; Supervised Machine Learning; Traffic Congestion