Title: Time series clustering based on prediction accuracy of global forecasting models
Authors: Angel Lopez Oriona - Universidade da Coruña (Spain) [presenting]
Jose Vilar - Universidade da Coruna (Spain)
Pablo Montero-Manso - Monash University (Australia)
Abstract: A novel method to perform clustering of time series is proposed. The procedure is based on the traditional K-means clustering algorithm and relies on two iterative steps: (i) K global forecasting models are fitted via pooling by considering the series pertaining to each cluster and (ii) each series is assigned to the group associated with the model producing the best forecasts according to a particular criterion. The resulting clustering partition contains groups which are optimal in terms of overall forecasting error and thus, the technique is able to detect the different prediction patterns existing in a given database. A simulation study shows that our method outperforms several alternative procedures concerning both clustering effectiveness and forecasting accuracy. The approach is also applied to perform clustering in three real-time series datasets.