Increasing number of clusters on fuzzy time series (fts) forecasting method

Abstract

Determining the effective number of clusters can influence forecasting results in FTS method’s applications. Unfortunately, the issue of how many clusters should be used to improve forecasting results has not been touched in previous researches. We observe for some different number of clusters and compare its Root Mean Squared Errors (RMSEs) results. The numerical simulation using Jakarta Composite Index (IHSG) data and the forecasting results show that RMSEs value decrease when the number of clusters is increased. The RMSEs value drops significantly when n = 50 clusters are used.