Predicting the Impact of Disruptions to Urban Rail Transit Systems

Service disruptions of rail transit systems become more frequent in the past decades in urban cities like Singapore, due to various reasons such as power failures, signal errors, etc. We study and predict the impact of disruptions to transit systems and commuters. This benefits service providers in making both short and long term plans to improve their services. Specifically, we define two metrics, stay ratio and travel delay, to quantify the impact. To tackle the main challenge of abnormal data scarcity, i.e., only 6 observed disruptions in our one-year data records, we propose to format the problem into a training problem on a feature space relevant to alternative route choices of the commuters. We demonstrate the new feature space corresponds to more similar data distribution among different disruptions, which is beneficial for training more generalisable predictors for future disruptions. We implement and evaluate our approach with a real-world transit card dataset. The result clearly shows that our method outperforms a range of baseline methods.