Knowledge and Systems Engineering, International Conference on (2011)
Oct. 14, 2011 to Oct. 17, 2011
DOI Bookmark: http://doi.ieeecomputersociety.org/10.1109/KSE.2011.10
I-k-Means is a popular clustering algorithm for time series data transformed by a multiresolution dimensionality reduction method. In this paper, we compare two different methods for initialization the I-k-means clustering algorithm. The first method uses kd tree and the second applies cluster-feature tree (CF-tree) to determine initial centers. In both approaches of clustering, we employ a new method for time series dimensionality reduction, MP_C, which can be easily made a multi-resolution feature extraction technique. Our experiments show that both initialization methods yield almost the same clustering quality, however the running time of I-k-Means initialized by using CF tree is a bit higher than that of the I-k-means initialized by using kd-tree. Both of the clustering approaches perform better than classical k-Means and I-k-Means in terms of clustering quality and running time.
Time series, Clustering, I-kMeans, Kd-tree, CF-tree
Nguyen Thanh Son, Duong Tuan Anh, "Two Different Methods for Initialization the I-k-Means Clustering of Time Series Data", Knowledge and Systems Engineering, International Conference on, vol. 00, no. , pp. 3-10, 2011, doi:10.1109/KSE.2011.10