MSTS: A System for Mining Sets of Time Series

Georg Lausen, Iztok Savnik, Aldar Dougarjapov

Abstract: A system to support the mining task of sets of time series is presented. A model of a set of time series is constructed by a series of classifiers each defining certain consecutive time points based on the characteristics of particular time points in the series. Matching a previously unknown series with respect to a model is discussed. The architecture of the MSTS-System (Mining of Sets of Time Series) is described. As a distinctive feature the system is implemented as a database application: time series and the models, i.e. series of classifiers, are database objects. As a consequence of this integration, advanced functionality as the manipulation of models and various forms of meta learning can be easily build on top of MSTS.

A preliminary version has been published in the Technical Report MSTS: A System for Mining Sets of Time Series.