Support vector regression based autoassociative models for time series classification

There are two paradigms for modeling varying length time series data, namely, modeling the sequence of feature vectors and modeling the sets of vectors. In this paper, we propose a regression based autoassociative model for modeling sets of vectors for time series data. We also propose a hybrid framework where a regression based autoassociative model is used for representing varying length time series data and then a discriminative model is used for classification. The proposed approach applied to speech emotion recognition task gives a better performance than the conventional methods.