Analysis and Prediction of Disruptions in Metro Networks

Abstract

Public transport disruptions can result in major impacts for passengers and operator. Our study objective is to predict disruption exposure at different stations, incorporating their location-specific characteristics. Based on a 13-month incident database for the Washington metro network, we successfully develop a supervised learning model to predict the expected number of disruptions, per type, station and time of day. This supports public transport authorities and operators to prioritize what type of disruptions at what location to focus on, to potentially achieve the largest reduction in disruption exposure. Our clustering results show that start/terminal and transfer stations are most susceptible to disruptions, mainly due to operations- and vehicle-related disruptions.