Predicting subway incident delays using text analysis based accelerated failure time model

Abstract

Subway delays would affect passengers’ travels and cause panic, having massive impacts on urban transit systems. Analyzing causes and the duration of subway delays would necessarily assist operational departments to take effective measures to alleviate the negative impact. Using Hong Kong subway incident data from 2005 to 2009, a text-based analysis of Accelerated Failure Time (AFT) model is developed to explore significant variables affecting subway delays and predict time lost as a result of subway delays. The results of this study suggest that a proposed mixture model outperforms the classical AFT model, because it utilizes topic model to extract candidate variables from textual information. In this way, factors affecting delays can be found. Beside the fact, that the results illustrate that non-device related factors i.e., human related factors, accidents and foreign object invasion cause longer delays. In conclusion, this study helped to develop predictions and thereby can reduce delays, minimizing the negative impact on the subway service and the passengers.