Development Of Subway Operation Accident Delay Models Using Maximum Likelihood Regression Tree Approaches

The breakdown of subway service could prolong the passengers" travel time, increase the probability of secondary accidents such as stampede accident and elevator accident and make it difficult for urban transportation administration. To predict subway operational accident delay accurately and timely could guide the passengers to reschedule their travel plan and lead the subway working staff to take measures for the delay. This paper aims at putting forward a model to recognize and evaluate the effecting factors producing subway accident delay. Moreover, this model could be used to predict subway operational accident delay. Firstly, this paper did literature review about the models and algorithms used for predicting and analyzing accident delays and durations in the field of traffic safety. It is concluded that, the tree methods based on regression analysis method could be used in subway operational accident delay analysis. The data comes from Hong Kong subway operation accident delay record. And the distribution was determined later. Secondly, according to the characteristics of the data, this paper came up with a model based on maximum likelihood regression trees aims to examine the relationship between subway accident delays and influencing factors. The goodness-of-fit results show that our model outperforms the traditional AFT models with fixed and random effects. And, the proposed model performs better in transferability. The main conclusions from this article are:tree-based model results are easy to read and understand; the presence of door failures appear most frequently, however it commonly produces short delays. Power failures can significantly increase the subway incident delay with a big variance. Longer subway delays are also associated with crash-involved and signal failures involved incidents. The subway accident delays on weekdays are generally shorter than those during weekends, especially during the peak periods. The proposed model of this study takes the advantage of parametric models and avoid the disadvantages of the non-parametric models such as instability and inaccuracy. The results are beneficial for subway engineers in proposing effective strategies to reduce subway accident delays, especially in the super large-sized cities with huge public travel demands.