Predicting Port-Catch Volume at Eastern Hokkaido Using Neural Networks

Abstract

Estimating the port fishing volume is an effective application to the fishery-industry information processing. Accurate prediction of port capture can help the transportation system operate more efficiently, reduce the time and cost of transportation, and contribute the freshness preservation of aquatic products. The LSTM (Long Short-Term Memory) neural network was introduced for the prediction of the catches of four ports of fisheries: Nemuro, Ochiishi, Habomai and Rausu, in the eastern area of Hokkaido from 2005 to 2015. And we newly used a designed model to solve the problem of sparse data. From the results, this model well works for solving the sparse-data problem by using a certain extent. Our proposed method becomes to be a kind of ICT development in the fishery industry.