摘要：Time series clustering has attracted increasing interest in the last decade, particularly for long time series such as those arising in the bioinformatics and financial domains. The widely known curse of dimension- ality problem indicates that high dimensionality not only slows the clustering process, but also degrades it. Many feature extraction techniques have been proposed to attack this problem and have shown that the performance and speed of the mining algorithm can be improved at several feature dimensions. However, how to choose the appropriate dimension is a challenging task especially for clustering problem in the absence of data labels that has not been well studied in the literature. In this paper we propose an unsupervised feature extraction algorithm using orthogonal wavelet transform for automatically choosing the dimensionality of features. The feature extraction algorithm selects the feature dimensionality by leveraging two conflicting requirements, i.e., lower dimensionality and lower sum of squared errors between the features and the original time series. The proposed feature extraction algorithm is efficient with time complexity O(mn) when using Haar wavelet. Encouraging experimental results are obtained on several synthetic and real-world time series datasets.
关键词：time series; data mining; feature extraction; clustering; wavelet
