Providing a Hybrid Clustering Method as an Auxiliary System in Automatic Labeling to Divide Employee into Different Levels of Productivity and their Retention

Abstract

Identifying productive employees and analyzing their turnover by data mining tools without human intervention is an attractive research field in human resource management. This study develops an innovative auxiliary system for automatic labeling of numerical data by providing a hybrid clustering algorithm of K-means and partition around medoids (PAM) methods to identify organizational productive employees and to divide them into different productivity levels. The model is evaluated by calculating the differences between actual and labeled values (93% labeling accuracy) and an innovative criterion for image processing of the final clusters using the singular value decomposition (SVD) algorithm. Ultimately, the results of the algorithm determine four labels of middle and good productive employees who leave the organization and excellent and weak productive employees who stay in the organization. According to each cluster, policies are adopted for their retaining, productivity improvement, replacement.