Rule-based granular classification: A hypersphere information granule-based method

As fundamental abstract constructs supporting the human-centered way of Granular Computing (GrC), information granules can be used to distinguish different classes of data from the perspective of easily understood geometrical structure. In this study, a three-stage rule-based granular classification method is proposed using a union of a series of hypersphere information granules. The first stage focuses on dividing each class of data into a series of chunks. The second stage concerns the construction of some hyperspheres around these chunks. These resulting hyperspheres form a union information granule to depict the key structural characteristics of the corresponding data through their union operation. At the final stage, the union information granules are refined and the rule-based granular classification model is emerged through using a series of “If-Then” rules to articulate the refined union information granule formed for each class with the corresponding class label. A number of experiments involving several synthetic and publicly available datasets are implemented to exhibit the advantages of the resulting classifier. The impacts of critical parameters on the performance of the constructed classifier are also revealed.