A machine learning approach to find the optimal routes through analysis of GPS traces of mobile city traffic

Abstract

The rapid urbanization in developing countries has modernized people’s lives in various aspects but also triggered many challenges, namely increasing carbon footprints/pollution, traffic congestion and high energy consumption. Traffic congestion is one of the major issues in any big city which has huge negative impacts, like wastage of productive time, longer travel time and more fuel consumption. In this paper, we aim to analyse GPS trajectories and analyse it to summarize the traffic flow patterns and detect probable traffic congestion. To have a feasible solution of the traffic congestion issue, we partition the complete region of interest (ROI) based on both traffic flow data and underlying structure of the road network. Our proposed framework combines various road features and GPS footprints, analyses the density of the traffic at each region, generates the road-segment graph along with the edge-weights and computes congestion ranks of the routes which in turn helps to identify optimal routes of a given source and destination point. Experimentation has been carried out using the GPS trajectories (T-drive data set of Microsoft) generated by 10,357 taxis covering 9 million kilometres and underlying road network extracted from OSM to show the effectiveness of the framework.