Comparative study of statistical and machine learning methods for streetcar incident duration analysis

Abstract

This study aims to investigate and identify the contributing factors to long-duration streetcar incident delay, where contingency plan could be activated. For comparative study, the performance of eight statistical and machine learning methods, including logistic regression model, Bayesian logit regression model, classification and regression tree model, K-nearest neighbours model, random forest model, gradient boosting model and artificial neural network model, have been compared and analysed based on the Toronto streetcar incident dataset in 2019 with 11418 streetcar incident records. The comparative study results show that the random forest method has the best performance, whose marginal effect analysis further demonstrates that the most significant contributing factors to streetcar incident delay duration are the morning peak period, the streetcar incidents types including mechanical failure, held by, diversion and late leaving the garage, as well as the month and weekday. The result of the paper could provide policy implication on timely streetcar incident clearance and contingency plan implementation.

Keywords:

• Streetcar incident
• long-duration delay
• comparative study
• statistical and machine learning methods
• marginal effect analysis

Disclosure statement

No potential conflict of interest was reported by the author.

Notes

1 Data source: https://ckan0.cf.opendata.inter.prod-toronto.ca/tr/dataset/ttc-streetcar-delay-data