SIFCM-Shape: State-of-the-Art Algorithm for Clustering Correlated Time Series

Abstract

Time-Series clustering is an important and challenging problem in data mining that is used to gain an insight into the mechanism that generate the time series. Large volumes of time series sequences appear in almost every fields including astronomy, biology, meteorology, medicine, finance, robotics, engineering and others. With the increase of time series data availability and volume, many time series clustering algorithms have been proposed to extract valuable information. The Time Series Clustering algorithms can organized into three main groups depending upon whether they work directly on raw data, with features extracted from data or with model built to best reflect the data. In this article, we present a novel algorithm, SIFCM-Shape, for clustering correlated time series. The algorithm presented in this paper is based on K-Shape and Fuzzy c-Shape time series clustering algorithms. SIFCM-Shape algorithm improves K-Shape and Fuzzy c-Shape by adding a fuzzy membership degree that incorporate into clustering process. Moreover it also takes into account the correlation between time series. Hence the potential is that the clustering results using this method are expected to be more accurate for related time-series. We evaluated the algorithm on UCR real time series datasets and compare it between K-Shape and Fuzzy C-shape. Numerical experiments on 48 real time series data sets show that the new algorithm outperforms state-of-the-art shape-based clustering algorithms in terms of accuracy.