Similarity Search in Time Series of Dynamical Model-based Systems

Abstract

Similarity search in time series is usually based on an assessment of the geometric similarity of the time series curves. In bioinformatics, dynamical model-based analysis and processing is used, where the curve itself is not meaningful. However, some internal features based on a model extracted from time series are meaningful. Therefore, the similarity is based on a dynamical model explaining the observation instead of being based merely on the superficial observation. There currently exist no methods for meaningful similarity search on such time series data emerging in bioinformatics. In this paper, we introduce a new similarity search method for time series based on similarity of internal features, called the perturbation method.