Cluster validation in k-Means clustering of mixed databases based on principal component analysis

Considering the close relation between k-Means clustering and principal component analysis (PCA), a cluster validation approach for k-Means partitions was proposed using analytical solutions of PCA. In this paper, the validation approach is further extended for handling mixed databases composed of not only numerical observations but also categorical observations. In the new validation approach for k-Means clustering of mixed databases, PCA solutions are given by considering optimal scaling of category observations, and the plausibility of k-Means solutions are evaluated by calculating deviations from the PCA solutions after Procrustean rotation.