Reducing the James–Stein Shrinkage Estimator for Automatically Grouping Heterogeneous Production Batches

Abstract

A reduction in the James–Stein shrinkage estimator might significantly increase the accuracy of cluster analysis of k-means for a relatively broad range of data. The efficiency of using the James–Stein shrinkage estimator for automatically grouping industrial products in homogeneous production batches is considered. Tests are conducted for batches of integrated circuits by comparing the shrinkage results with those obtained using the traditional k-means algorithm. The dataset is normalized according to the values of the acceptable drift, acceptable parameters, and standard deviation. As established using the Rand index, clustering is far more accurate in the automatic grouping of industrial products in homogeneous production batches, when average values of inconclusive parameters drop to zero. It is established that the reduction of the James–Stein shrinkage estimator decreases the influence of inconclusive parameters of standard data to acceptable values.