Mining Time Series Data: A Selective Survey

Abstract

Time series prediction and control may involve the study of massive data archive and require some kind of data mining techniques. In order to make the comparison of time series meaningful, one important question is to decide what similarity means and what features have to be extracted from a time series. This question leads to the fundamental dichotomy: (a) similarity can be based solely on time series shape; (b) similarity can be measured by looking at time series structure. This article discusses the main dissimilarity indices proposed in literature for time series data mining.