A Regime Switching Independent Component Analysis Method for Temporal Data

A mixture of independent component analysis method for temporal data is presented in this paper. The method is derived by modeling the observations as a mixture of ICA (mICA). mICA model has been applied to data classification and image processing. However, it is hard to use mICA in assigning class memberships of temporal data. In the proposed method, memberships of the data are modified according to its past values in the learning process. It shows that the proposed method is able to detect the switch between mixtures in highly overlapped data, which have smaller error than traditional mICA method.