A novel method for time series symbolization based on singular event features clustering

This Qualitative abstract representation of time series is a precondition of pattern discovery. A novel method for time series symbolization based on singular event features clustering is proposed in this paper. The first step of it is to extract singular event features based on multi-scale wavelet which can divide time series into event sequences with independent trend. Secondly, cluster the events that represented by transform parameters through the novel fuzzy immune genetic algorithm to implement symbolization. Each event is identified by the cluster it belongs to. The proposed method is applied to unstable financial time series symbolization. The proposed method can be used to discover significant similar patterns, classification and associated patterns from time series.