A multilevel distance-based index structure for multivariate time series

Multivariate time series (MTS) datasets are common in various multimedia, medical and financial applications. In previous work, we introduced a similarity measure for MTS datasets, termed Eros (extended Frobenius norm), which is based on the Frobenius norm and principal component analysis (PCA). Eros computes the similarity between two MTS items by measuring how close the corresponding principal components (PCs) are using the eigenvalues as weights. Since the weights are based on the data items in the database, they change whenever data are inserted into or removed from the database. In this paper, we propose a distance-based index structure, Muse (Multilevel distance-based index structure for Eros), for efficient retrieval of MTS items using Eros. Muse constructs each level as a distance-based index structure without using the weights, up to z levels. At the query time, Muse combines the z levels with the weights, which enables the weights to change without the need to rebuild the index structure. In order to show the efficiency of Muse, we performed several experiments on a set of synthetically generated clustered datasets. The results show the superiority of Muse as compared to sequential scan and M-tree in performance.