Host: Japan SOciety for Fuzzy Theory and intelligent informatics
Co-host: The Korea Fuzzy Logic and Intelligent Systems Society, IEEE Computational Intelligence Society, The International Fuzzy Systems Association, 21th Century COE Program "Creation of Agent-Based Social Systems Sciences"
Independent Component Analysis (ICA) is a technique for blind source separation and is also useful in regression (prediction) task when only a subset of random variables is observed. Because the task of regression by ICA consists of finding non-Gaussian components, it is closely related to projection pursuit regression. Local independent component analysis (Local ICA) is a non-linear extension of linear ICA models that extract local feature values by applying linear ICA in conjunction with suitable clustering algorithms. This paper proposes a switching regression model, in which local linear structure is first captured by fuzzy clustering, and then a non-linear regression model is estimated by a modified ICA model considering fuzzy memberships in each cluster.