Optimized Multi-Algorithm Voting: Increasing objectivity in clustering. (15th March 2019)

• Record Type: Journal Article
• Authors: Kempen, Regina; Meier, Alexander; Hasche, Jens; Mueller, Karsten
• Is Part Of: Expert systems with applications. Volume 118(2019)
• Page Start: 217
• Page End: 230
• Publication Date: 2019-03-15
• Subjects: Clustering -- Integrative methods -- Multi-algorithm voting -- Work-related values; Expert systems (Computer science) -- Periodicals; Systèmes experts (Informatique) -- Périodiques; Electronic journals; 006.33
• DOI: 10.1016/j.eswa.2018.09.047
• Languages: English
• ISSNs: 0957-4174

Abstract

Highlights: Depending on clustering algorithm used, the results may change. Based on Multi-Algorithm Voting, robustness of cluster solutions can be improved. OMAV is developed that uses an optimization algorithm as integrative method. OMAV is applied to the example of country clustering using GLOBE data. Increased robustness and reduced subjectivity are demonstrated. Abstract: Currently, the influence of a single statistical cluster algorithm on the results of clustering procedures represents a major threat to the objectivity in clustering. To exemplify this question, this paper refers to country clustering in cross-cultural research. In this field, previous research has determined differing numbers of clusters, depending on choices available for the clustering procedure, leading to a high number of inconsistent results. Hence, it is argued that the variety in cluster solutions induced by the choice of different statistical cluster algorithms should be reduced. To this end, this study builds on Multi-Algorithm Voting (MAV) procedure introduced by Bittmann and Gelbard (2007) and presents an advancement to the MAV method. Specifically, MAV procedure is refined for the analysis of larger data sets using the simulated annealing algorithm for optimization. The use of this Optimized MAV (OMAV) is then demonstrated for country clustering in cross-cultural research. Specifically, a set of 57 countries is divided into 12 clusters based on work-related values obtained from GLOBE database reported in House et al. (2004). Thus, results clearly show that the objectivity of clustering results can be significantly improved based on OMAV. Implications for expert and intelligent systems on the use of OMAV are discussed. Namely, OMAV represents a powerful tool supporting the decision-making process in cluster analysis reducing the number of subjective and arbitrary decisions. Taken together, this study contributes to existing literature by providing an integrative and robust method of country clustering using OMAV and by presenting country clusters applicable to various settings.