Time series clustering using numerical and fuzzy representations

In this paper a Fuzzy Behavior Clustering approach for time series clustering that combines fuzzy techniques with well-known clustering algorithm is presented. The proposed time series clustering approach has three stages. At the first one time series model based clustering using fuzzy techniques (F-transform and a general fuzzy tendency) is proposed. As a result, quantitative and linguistic representations of the time series model are derived. Such representations allow us to group the time series with the similar patterns of an additive model and therefore with the same types of a behavior. A feature extraction and a point based time series clustering are used at the subsequent stages for more detailed data splitting in a larger number of clusters. The experiments showed the improvement of the time series clustering results using the proposed approach.