PREDICTION OF FERRY ARRIVAL TIMES USING MACHINE LEARNING
Mehmet Duman University of Massachusetts, Dartmouth MattDuman7@gmail.com
Faculty Advisor Dr. David Koop
In this project, we study machine learning methods to analyze publicly available data for Washington State Ferries’ vessel trajectories. We studied both on-time and delayed trips, generated predicted arrival times, and compared our predictions with Washington State Ferries’ own estimate time of arrival (ETA). The project used Python and the SciKit-Learn (sklearn) library which contains a variety of algorithms that can generate models and evaluate their predictive ability. The main contributions of the project are an analysis of the data retrieved from WSF website for the first quarter of 2017 and the predictions derived from different models created using linear regression, k-nearest neighbors, and random forest regression. We compared the results of the models to discover which method produces the best predictions in addition to comparing with the posted ETA prediction. Our study shows both that WSF’s own ETA prediction can have significant errors, and that random forest regression gives the most accurate prediction of all the models tested.