Fuzzy time series prediction using hierarchical clustering algorithms

In many cases, the k-means clustering algorithm has been most frequently used to the field of data mining, fuzzy control systems and prediction since it was designed in simple procedures and excellent ability of classification. However, it sometimes brought about the failed results for non-linear data by classification behavior caused by just considering the statistical characteristics of non-linear data such as distances between data. To overcome the problems above, this paper proposes a new clustering algorithm of which the structure hierarchically classifies non-linear data. The proposed hierarchical classification technique consists of two levels, called upper clusters and lower fuzzy sets, using the cross-correlation clustering algorithm combined with the k-means clustering algorithm (HCKA), and it was able to improve classification accuracy. In addition, this paper constructs multiple model fuzzy predictors (MMFPs) corresponding to difference data of original time series, which was able to reflect the various characteristics of the time series to the proposed system. Simulation results show that the proposed system was effective and useful for modeling and predicting non-linear time series.