Classification of multivariate time series using locality preserving projections

A Feature Extraction Method for Multivariate Time Series Classification Using Temporal Patterns

“…Li et al [7] proposed a feature extraction method, Singular Value Decomposition (SVD), to reduce the different length of data to feature vectors, and then apply SVM on the feature vectors to classify MTS data. Weng et al [6] project original MTS into PCA subspace by throwing away the smallest principal components firstly, and then MTS samples in the PCA subspace are projected into a lower-dimensional space by using supervised Locality Preserving Projection (LPP). However, the above existing extraction method may lose the dependency relationship information among different univariate time series.…”

“…A number of different approaches have been proposed for univariate time series classification [9][10][11], however, few papers are found about multivariate time series classification in the literature [6].…”

Enhancing learning algorithms to support data with short sequence features by automated feature discovery

“…Numerous approaches of preprocessing and manipulating vector and sequence data to a list of discrete values exist. For example, sequence mining can efficiently search for patterns that are correlated with the target classes [33]; classification of multivariate time series using locality preserving projections that extract the features by projecting the vector samples onto a lower dimensional space in which the multivariate time series samples related to the same class are close to each other [49]; and multivariate time series classifier, which adopts triangle distance function as similarity measure extracts some meaningful patterns from an original data and uses traditional machine learning algorithms to create a classifier based on the extracted patterns [54]. Existing approaches preprocess series data before classification.…”

Selective Subsequence Time Series clustering

“…Time series clustering [2,6,15,13,24] is one of the most popular tasks in time series data mining community [5,16,18,25,26,20]. Most algorithms generally perform whole time series clustering [24,15].…”