DOKTORI (PhD) ÉRTEKEZÉS

Correlation-based elastic similarity measures for data mining of technological time series In the last decades, the availability of easily accessible computational resources and storage capacities has made it possible to apply classical data mining techniques on time series data. Such methods have rapidly gained popularity in engineering applications, thanks to their ease of use, effectiveness and the fact that several time independent problems can easily be formulated and solved as a time series related task. Over the time, however, it became clear that dissimilarity measures inherited from classical data mining methods are not well-suited for time series data, because they do not take into account the nonlinear fluctuation of the time axis. To overcome this problem, elastic dissimilarity measures were proposed; however, the expected good results still could not be achieved in several applications. The aim of this thesis is to provide solutions for time series similarity problems that previously could not be handled with elastic dissimilarity measures. Three novel similarity measures built around dynamic time warping are presented, which enable more precise data mining of time series by incorporating new aspects. The key concepts of the first measure, correlation-based dynamic time warping, are that homogeneous segments of the multivariate time series can be created from correlation point of view using principal component analysis motivated segmentation, and the distance of these segments can then be computed using any principal component analysis-based similarity measure. This way, correlation-based dynamic time warping enables comparison of multivariate time series not only by considering the shifts of the time axis but also the alternation of the correlation structure itself. Driven by this solution, the second similarity measure — while it also takes advantage of the elasticity provided by dynamic time warping — replaces principal component analysis with its dynamic version in correlation-based dynamic time warping, and thus it enables comparison of multivariate time series based on the changes in process dynamics. Finally, a similarity measure that combines two different features is presented for piecewise linear approximated time series. It demonstrates that the application of