Cluster Analysis from a Game Theoretical Framework

Abstract: Although regression analysis is the predominant method of statistical analysis in econometrics, unsupervised methods such as cluster analysis can bring novel insight into detailed data sets. In this paper, we analogize game theory with cluster analysis, and find that the k-means algorithm is a Nash equilibrium in a repeated game framework. Along with the clustering process, we also present dimensionality reduction methods in a game theoretical framework. We then present the results of cluster analysis on corporate data from the mobile game company King. Our results show that cluster analysis can extract valuable patterns from large data sets in both static and time-series data by segmenting the population into groups with overarching similarities, rather than forking on various combinations of variables.