An unsupervised method to determine the optimal number of independent components

Abstract In this paper, we present a new method to determine the optimal number of independent components after applying an Independent Component Analysis ( ICA ) to a set of mixed signals. The proposed method, called Linear Correlations between Components ( LCC ), uses the JADE algorithm to calculate the independent components. The LCC method allows to automatically select the optimal number of independent components in an unsupervised way without any previous knowledge. It has been tested using synthetic mixed signals where the number of pure (or independent) signals is known. This method is very simple, fast and easy to implement.