Predicting Traffic Flow Propagation Based on Congestion at Neighbouring Roads Using Hidden Markov Model

Nowadays traffic congestion has become significantly worse. Not only has it led to economic losses, but also to environmental damages, wastage of time and energy, human stress and pollution. Generally, traffic congestion is a ripple effect of a road congestion on neighboring roads. When congestion occurs, it will propagate through the road network due to increasing traffic flow. One of the complexities of traffic congestion is unpredictability, thus it is difficult to represent traffic flows by numerical equations. One possible approach is to use the spatial historical data of traffic flow and relate them with traffic condition (congestion or clear) using statistical approach. Studies on traffic flow propagation generally involves visualization with real time GPS trajectory data to help analyze traffic flow propagation using human vision. Our research focuses on traffic flow pattern based on data from sensors without having information about the connected roads. We study spatial and temporal factors that influence traffic flow near a congested road in a neighboring area. Hence, our study investigates the relationship of roads in a neighboring area based on the similarity of traffic condition. Roads with high relationship with other neighbouring roads are identified by extracting spatial and temporal features using traffic state clustering. Grey level of co-occurrence matrix (GLCM) is utilized with spectral clustering to cluster road segments that have the same duration of road congestion in terms of day and time intervals. The emission probability is then calculated for prediction of traffic state impact of road congestion in neighboring area using Hidden Markov Model (HMM). We proposed HMM together with our clustering method to predict traffic state impact of road congestion. The experimental results show that the accuracy of prediction using the proposed HMM achieve 89%