Real-time congestion detection using public transport data

Abstract

Congestion is one of the biggest problems which affects life quality and has an impact on social and economic conditions. Tackling traffic congestion has always been a challenge. The situation is more aggravated for developing countries, like India, due to their huge population and overutilization of basic resources. Existing solutions in terms of policies like the construction of new flyovers, underpasses, widening of roads have failed miserably. Prevalent technical solutions for congestion detection like deployment of the camera, RFID and other sensors are quite popular but they are more popular in developed countries. Developing countries are not much readily involved in such strategies as they require economic contributions and maintenance. Also web mapping service providers like Google, Bing, HERE uses crowdsourced information of mobile devices through their applications. Such kind of private data is not available to the government agencies which they can utilize to improve their transport system and solve congestion problems. So we study the effectiveness of congestion detection if only public transport data is available. We investigate the utility of the real-time bus Spatio-temporal data, which is sparse and has missing values for the task of congestion detection. Such a system if works with good accuracy, it would make government authorities not to depend on private players. We provide a real-time congestion detection mechanism that exploits GPS sensors installed on Delhi’s DIMTS cluster buses to provide fast & reliable congestion status. We compare multiple strategies and observe 70% f1 score and 80% recall at best by a simple statistical-based method. We also analyze this data for the application of hotspot detection and identifying popular bus stops.