An efficient and accurate optimization method of sliding window size for PAA

Abstract

PAA is an important algorithm in time series dimensionality reduction. However, how to determine the sliding window keeps an open issue for PAA and its derivatives. In this paper, a new optimization method to decide the PAA window is proposed based on root mean square distance measure. A rate of information loss is proposed to overcome the scalability issue, which can be used to balance information loss and query performance improvement caused by PAA transformation. Experiment results with a real time series dataset demonstrate that the method is effective and feasible to determine the PAA window size and optimize the whole performance of PAA algorithm.