Feature-based time-series classification

The paper is devoted to the multiclass time series classification problem. The feature-based approach that uses meaningful and concise representations for feature space construction is applied. A time series is considered as a sequence of segments approximated by parametric models, and their parameters are used as time series features. This feature construction method inherits from the approximation model such unique properties as shift invariance. The authors propose an approach to solve the time series classification problem using distributions of parameters of the approximation model. The proposed approach is applied to the human activity classification problem. The computational experiments on real data demonstrate superiority of the proposed algorithm over baseline solutions.