Dynamic Bus Arrival Time Prediction exploiting Non-linear Correlations

Public transport buses exhibit uncertainties in their arrival/travel times, due to several factors such as signals, dwell times at bus stops, seasonal variations, fluctuating travel demands etc. Further, factors like excess vehicles, acute lack of lane discipline and diversity in modes of transport additionally plague the traffic in the developing world. Owing to these diverse factors, the bus arrival time prediction (BATP) continues to be a challenging problem especially in the developing world. The current work proposes a data-driven method to perform bus arrival time prediction in real-time. Unlike existing approaches, the proposed method explicitly learns both the spatial and temporal correlations/patterns of traffic in a novel and general fashion. In particular, it first detects the unknown order of spatial dependence and then learns general nonlinear and non-stationary correlations using supervised learning, for this detected order. The real time prediction problem is now posed as an inference problem on an associated non-linear dynamical system (NLDS) model. We propose to use an Extended Kalman Filter (EKF) to solve this inference problem in an approximate but efficient manner. We demonstrate the effectiveness of our proposed algorithm on real field data from challenging traffic conditions in the developing world. Our experiments demonstrate that the proposed EKF outperforms diverse existing state-of-art data driven approaches proposed for the same problem.