From numeric data to information granules: A design through clustering and the principle of justifiable granularity

Designing information granules used intensively in Granular Computing is of paramount relevance to the fundamentals of the discipline. Information granules are key functional components in granular models, granular classifiers, and granular decision-making models. The design of information granules is central to the discipline of Granular Computing. In this study, we introduce a way of designing information granules by combining the mechanisms of unsupervised and supervised learning and subsequently using the principle of justifiable granularity. An overall design process consists of two phases. First, the granulation process involves hierarchical clustering or K-means clustering. It is followed by a parametric refinement of information granules realized by the principle of justifiable granularity. The characterization of information granules is offered in terms of measures of coverage, specificity, and entropy. Experimental results including synthetic data and publicly available data are covered to demonstrate the performance of the proposed approach.