Abstract: Over the past few decades, service disruptions of rail transportation systems have increased in major cities for a number of reasons, such as power outages, signal issues, etc. The impacts of disruptions on users and transit networks are studied and projected. This makes it easier for service providers to set both short- and long-term goals to enhance their offerings. We precisely establish two metrics—stay ratio and journey delay—to assess the impact. In order to overcome the main challenge of unusual data scarcity—namely, the fact that there were only 6 documented disruptions in our one-year data sets—we propose structuring the issue as a training problem on a feature space relevant to alternate commuter route choices. We demonstrate that the new feature space correlates to more comparable data distribution across different disruptions, which is helpful for creating disruptor predictors that can be used more widely. We test and evaluate our approach using a dataset from real transit cards. The result clearly shows that our strategy performs better than a variety of benchmark techniques.
Keyword: Service disruption, impact prediction, data scarcity
Copyright Form