Title: Dynamic clustering of multiple multivariate time series: Application to climate data
Authors: Ceylan Yozgatligil - Middle East Technical University (Turkey) [presenting]
Sipan Aslan - middle east technical university (Turkey)
Cem Iyigun - middle east technical university (Turkey)
Abstract: A new time series clustering approach which is a nonlinear time series model based dynamic clustering approach for multiple multivariate time series is proposed. Observing similarity-dissimilarity between time series is accomplished by evaluating the approximations to their unknown data generating mechanisms rather than comparing trace-like pattern similarities in the time series. In other words, the proposed approach is mainly aimed at forming and acquiring distinguishing information-features throughout the given period of a time series using the threshold autoregressive and threshold vector autoregressive models where the actual nature of the underlying process or data generating mechanism is not known. The effectiveness of the proposed approach is illustrated via simulation examples, and then, it is applied to the problem of defining climate regions based on multivariate meteorological time series.