Proposal of Percolative Learning Method for the Baltic Dry Index Forecasting

Abstract

Japan depends on imports for almost all resources, which are supported by maritime trade. Since the shipping industry is a single global market and highly competitive, it is important to anticipate future fluctuations for stable transportation. On the other hand, existing research on time-series forecasting uses only past observations, making it difficult to predict future fluctuations more accurately. In this paper, we propose a training method of percolative learning model. We apply this method to test problems of predicting future shipping market. The results indicate that the proposed method is more accurate and effective than the conventional method.