An efficient similarity search for multivariate time series

Abstract

In order to efficiently perform similarity search for Multivariate Time Series (MTS)datasets, a distance-based index structure (Dbis) for similarity search was presented. The dimension of MTS database was reduced firstly by Principal Component Analysis (PCA). The principal component of MTS was parted by cluster, and a MTS item was selected as reference point from each partition. The MTS items in each partition were transformed into a single dimensional space based on their similarity with respect to a reference MTS item. This allowed the MTS items to be indexed by using a B+-tree structure. An extended Frobenius norm (Eros) was used to compare the similarity between MTS items. Several experiments on a financial MTS database were performed. The results show the effectiveness of Dbis.