Research on similarity measure for time series based on SAX
基于SAX的时间序列相似性度量方法*

Symbolic approximation is an effective dimensionality reduction technique for time series, its similarity measure is a basis for various mining tasks. MINDIST_PAA_iSAX is a distance function based on symbolic aggregate approximation (SAX), but it does not satisfy symmetry, so it has limitation in mining time series. This paper put forward and proved a symmetric distance measure Sym_PAA_SAX to be lower bounding to Euclidean distance. Experiments on real and synthetic data sets show its better tightness of lower bounding and lower false positives rate in similarity search.