Multivariate time series clustering research based on Hadamard transform
Fang Xiao-zhao · Jisuanji gongcheng yu sheji · 2012
A effective clustering method is presented based on researching the present clustering algorithm.Firstly,a multivariate time sequence is reduced-order by discrete Hadamard transform.Then,the eigenvalue of multiple variables related coefficient matrix are the weights of multiple time series transformation matrix.Finally,Using the matrix similarity measure with weights,improved K-means algorithm is used to cluster the multivariate time series.Experimental results show this method can achieve multivariate time series clustering effectively.