MULTI-OBJECTIVE GENETIC MOTIF DISCOVERY TECHNIQUE FOR TIME SERIES CLASSIFICATION

Abstract

Time series is a sequence of continuous and unbounded group of data observations recorded from many applications. Time series motif discovery is an essential and important task in time series mining. Discovering motifs in time series has attracted the researcher's attention for efficient time series classification problems and several algorithms have been proposed to solve the problem. However, these algorithms depend on predefined parameters like support, confidence, and length of the motif and they are sensitive to the parameters. To overcome the challenge, this paper proposes a multi-objective genetic algorithm to discover a good trade-off between representative and interesting motif. The discovered motifs are validated for their potential interest in time series classification using nearest neighbour classifier. Extensive experiments show that the proposed approach can efficiently discover motifs with different lengths and more accurate than state-of-the-art time series techniques. The paper also demonstrates the efficiency of motif discovery in classifying the large time series medical data from MIT-BIH-arrhythmia database.