An Absolute Differences K-Means Clustering Approach on Determining Intervals to Optimize Fuzzy Time Series Markov Chain Model

Abstract

Fuzzy Time Series models have been developed in various ways, one of which is determining the intervals. Several methods were applied to determine the intervals, but the performances are still not optimal. This paper proposes a new approach that uses a combination of Absolute Differences and K-means Clustering in the Fuzzy Time Series Markov Chain model. K-means Clustering made the interval more flexible and compact based on the data it clustered. In addition, Absolute Differences was used as the based method to define how many intervals to be made. This study used Taiwan Capitalization Weighted Stock Index (TAIEX) as benchmark data to evaluate the proposed method, which produced an average Mean Absolute Percentage Error (MAPE) value of 0.42, and an average Root Mean Squared Error (RMSE) value of 51.09 for the test data. The proposed method outperformed other compared researches at the end of this paper in terms of prediction accuracy.