Although clustering techniques or methods have been extensively applied to find groups in data, they ignored the input-output relationship. The social science applications mainly revolve around the relationship between the input and output of attribute values. Without these special relationships, clustering results may suffer from a biased casual relationship among feature items. To consider the input-output relationships, this study uses the concept of stratification data envelopment analysis (DEA) to group data points according to their efficiency performance scores based on the context–decision analysis. The stratification DEA clustering approach could be applied not only to clustering but also to illustrating the group preference. Three criteria, clustering criterion, balancing criterion and benchmarking criterion, are used to validate the quality of clustering. In this study, the stratification DEA clustering approach, the feature-based clustering method and the scale-based clustering approach are compared based on these three criteria.