Multiple classifiers for time series classification using adaptive fusion of feature and distance based methods

Time series classification is a supervised learning problem used in many vital applications. Classification of data varying with time is considered an important and challenging pattern recognition task. The temporal aspect and lack of features in time series data makes the learning process different from traditional classification problems. In this paper we propose a multiple classifier system approach for time series classification. The proposed approach adaptively integrates extracted local and global features together with distance similarity based methods. A feature extraction process is performed, followed by training of base classifiers using different features and then decision fusion of single classifiers. Also in this study, an evaluation of single and multiple classifiers is performed on different combinations of extracted features. Dynamic and static methods for decision fusion are explored. We investigate character and sign language recognition applications. Our proposed method is tested against various single and multiple classifiers trained using different feature spaces. The results for the different training schemes are presented. We demonstrate that the proposed model produces better accuracy in most cases.