Slopewise Aggregate Approximation SAX: keeping the trend of a time series
Lamprini Pappa, Petros S. Karvelis, George Georgoulas, Chrysostomos D. Stylios · 2021 IEEE Symposium Series on Computational Intelligence (SSCI) · 2021
In this work, we introduce the Slopewise Aggregate Approximation (SAA), an innovative variation of the Piecewise Aggregate Approximation. The Slopewise Aggregate Approximation (SAA) is used as a novel core step for the Symbolic Aggregate Approximation method. SAA efficiently describes the trend at a time series signal since it incorporates information regarding the shape and fluctuation of the time series while simultaneously achieving the problem's dimensionality reduction. Then, by applying the discretization technique, the problem is transformed into a symbolic space problem, and the Intelligent Icons are the features that come out and feed a Near Neighbour classifier for a Human Activity Recognition problem. The results achieved by the proposed method are directly compared to two relevant past implementations and exhibit a considerable increase in classification metrics.