Maximum likelihood regression tree with two-variable splitting scheme for subway incident delay

Considering possible variable interaction effects, this study develops a maximum likelihood regression tree-based (MLRT) model using the proposed two-variable splitting method to describe subway incident delays. A MLRT comprising 13 leaf nodes is built with Hong Kong subway incident data from 2005 to 2012 and a log-logistic distributed accelerated failure time (AFT) model is developed separately for each leaf node. The comparison of model performance indicates that our developed model outperforms traditional AFT models and the tree-based model building based on the traditional single-variable splitting scheme. The probability of subway incident delay being unacceptable can be predicted using our developed model, which can be utilized as a basis for alerting commuters to the necessity of rescheduling their trips in the event of a subway incident.