Fuzzy local independent component analysis with external criteria and its application to knowledge discovery in databases Original Research Article

Local feature values derived by hybrid approaches to fuzzy clustering and multivariate data analysis have been used for knowledge discovery in databases (KDD). They often, however, fail to reveal intrinsic structure because observed variables are easily influenced by external variables. This paper proposes an enhanced technique of local independent component analysis (Local ICA), which extracts independent components uncorrelated to some external criteria. The new technique is applied to knowledge discovery from POS transaction data with the goal of the analysis being to reveal the relationship between the number of customers and days of the week.