Granular rule-based modeling using the principle of justifiable granularity and boundary erosion clustering

Rule-based models constructed by “IF-THEN” fuzzy rules are commonly used in a complex and nonlinear system. In this study, a novel modeling method is established to generate fuzzy rules based on experimental evidence. Such modeling is realized by utilizing the boundary erosion algorithm to cluster the input samples and the principle of justifiable granularity to granulate the corresponding output. To further examine the performance of the designed rule-based model under different granularity levels, a model with the finer information granules is designed for rule extraction in each cluster. The proposed models are assessed on the synthetic and ship datasets, where the comparison between the granular output and the original data value is considered as the evaluation metric based on the converge and specificity of information granules. Numerical results show that the rule-based models, which incorporate information granules to form representative rules, perform better in analyzing the structure of the arbitrary-shaped datasets and offer a potential application in ship management.