Feature extraction from POS transaction data by using local independent components

摘要

Independent component analysis (ICA), which has been developed mainly in signal processing, is a useful technique for Projection Pursuit as will. For some nonlinearly distributed data, the data set is partitioned into several groups using Fuzzy c-Varieties clustering method before applying the ICA algorithm, which constitute a fuzzy version of Fast ICA by Hyvarinen et al.. This paper discusses feature extraction from local independent components by applying the techniques to POS (point-of-sales) transaction data.