Identifying the root cause in urban road networks and ranking the influential factors can benefit traffic management for improving traffic condition. Traditional congestion identification studies paid attention to identify traffic bottlenecks, namely the most vulnerable points in a road network, without consideration of root causes that leading to the congestion. In this paper, we propose a gradient boosting decision trees (GBDTs) based method to identify the root cause of road network congestion and rank the influential factors using different types of explanatory variables. Based on Sioux Falls network, different signal control strategies at intersections and number of lanes on road segments under different traffic flows are conducted as samples using Simulation of Urban Mobility (SUMO) to train and test the GBDT model. Simulation results indicate that the GBDT model can achieve superior performance in average travel speed prediction and identify the root causes of congestion by prioritizing the relative importance of influential factors, such as lane numbers and signal control strategies, compared with other algorithms.
Root Cause Identification for Road Network Congestion Using the Gradient Boosting Decision Trees
2020-12-01
441328 byte
Conference paper
Electronic Resource
English
Wiley | 2024
|