A multi-act sequential game-based multi-objective clustering approach for categorical data

Abstract

Clustering categorical data, where no natural ordering can be found among the attributes values, has started drawing interest recently. Few clustering methods have been proposed to satisfy the categorical data requirements. Most of these methods have focused on optimizing a single measure, however, several applications in different areas need to consider multiple incommensurable criteria, often conflicting, during clustering. Motivated by this, we developed a multi-objective clustering approach for categorical data based on sequential games. It can automatically generate the correct number of clusters. The approach consists of three main phases. The first phase identifies initial clusters according to an initialization mechanism which has an important effect in the final clustering result. The second phase uses multi-act multi-objective sequential two-player games in order to determine the appropriate number of clusters. A methodology based on backward induction is used to calculate a pure Nash equilibrium for each game. Finally, the third phase constructs homogenous clusters by optimizing intra-cluster inertia. The performance of this algorithm has been studied on both simulated and real-world datasets. Comparisons with other clustering algorithms illustrate the effectiveness of the proposed approach.