Efficiency of distance measures in the automatic grouping of electronic radio devices by k-means algorithm

Abstract

This study investigates the problem of separation mixed lot of semiconductor devices for space applications into homogeneous production batches. The impact of distancemeasures in clustering algorithmk-means on the accuracy of the separating a mixed production batch is studied. This paper proposes the K-means algorithm withsquared Mahalanobis distance as an alternative the squared Euclidean distance. With the various combined mixed lots of semiconductor devices it was shown, that the proposed squared Mahalanobis distance highly depends on the method for determining initial cluster centers.