Real-Time Bayesian Micro-Analysis for Metro Traffic Prediction

Metro transport plays a large role in major cities around the world as an easily accessible and convenient means of transit. We propose a novel approach to forecast the metro network flow of passengers, which is exceptionally useful for city planning. For instance, accurate estimations of passenger outflow provide valuable insight in deciding where and when to add new trains and stations. We present a micro-prediction approach to predict individual passenger's destination station and arrival time. As a global apriori model we empirically learn a probability distribution of origin-destination (OD) station-pairs using analysis on historical data and estimate travel times between stations. Then, we condition the OD probability distribution by the current travel time of an individual passenger using Bayesian learning. For each station, the summation of the probability distribution of each passenger in the network produces the expected outflow. Our experimental evaluation shows that our model outperforms baseline approaches, thus showing that our model can be successfully implemented for a wide array of passenger traffic flow data for smart city planning.