The staggering amount of streaming time series coming from the real worldcalls for more efficient and effective online modeling solution. For timeseries modeling, most existing works make some unrealistic assumptions such asthe input data is of fixed length or well aligned, which requires extra efforton segmentation or normalization of the raw streaming data. Although someliterature claim their approaches to be invariant to data length andmisalignment, they are too time-consuming to model a streaming time series inan online manner. We propose a novel and more practical online modeling andclassification scheme, DDE-MGM, which does not make any assumptions on the timeseries while maintaining high efficiency and state-of-the-art performance. Thederivative delay embedding (DDE) is developed to incrementally transform timeseries to the embedding space, where the intrinsic characteristics of data ispreserved as recursive patterns regardless of the stream length andmisalignment. Then, a non-parametric Markov geographic model (MGM) is proposedto both model and classify the pattern in an online manner. Experimentalresults demonstrate the effectiveness and superior classification accuracy ofthe proposed DDE-MGM in an online setting as compared to the state-of-the-art.
URL
Affiliations
Abstract
Translation (by gpt-3.5-turbo)
Summary (by gpt-3.5-turbo)