Similarity Search in Multidimensional Time Series using the Coulomb’s Law

Abstract

With the technological advancement, associated to the low production cost of instruments used to collectobservations, there has been an increase in the amount of available data for analysis. The collected data present intrinsicrelations between them that are not perceptive without a careful analysis, requiring the use of specic techniques tomanipulate them. In this context, we propose a time series descriptor, based on the principle of the Coulomb Law, toperform similarity search over multidimensional time series. The proposed descriptor is composed of a new time seriesextractor and a new distance function for multidimensional time series. Moreover, this paper presents the Coulombmethod that describe how to employ the proposed descriptor to perform similarity search over multidimensional timeseries. According to the experiments performed over climatic and medical databases, the proposed method promotes alarger reduction in the feature vector size and achieves higher accuracy values when compared with other time seriesdescriptors based on Fourier Transform and Brute-force solution.