Prediction of Liner Shipping Operations by Machine Learning

Abstract

A liner ship is the primary means of access to Tobishima, a remote island in Sakata City, Yamagata Prefecture. The decision to depart is made on the morning of the day based on weather data such as wind speed and wave height in the surrounding area. Therefore, it is difficult to make a plan until the day of departure, and there is a need for predicting the operation of liner ships up to the day before. In this study, we investigate the prediction of liner ship operations by machine learning such as SVM and CNN using AMeDAS data of Japan Meteorological Agency and satellite images. As a result, we have confirmed that the prediction accuracy is about 90% on the same day and about 80% on the next day and that the prediction accuracy is about 90% even one week ahead of the departure.