Abstract
A deterministic procedure for k-Means clustering was proposed considering the close connection with PCA solutions, in which a relaxed solution was derived based on the optimality of the k-Means objective function. Although the relaxed solution is a rotated one and the rotation matrix is unknown, it can be used for cluster validation of k-Means solutions in conjunction with Procrustean transformation of principal component scores. In this research, it is demonstrated that the PCA-guided approach is still useful for data with missing values and is also applicable for cluster validation in case of incomplete data.
