Abstract
In the study, we propose a concept of incremental fuzzy models in which fuzzy rules are aimed at compensating discrepancies resulting because of the use of a certain global yet simple model of general nature (such as e.g., a constant or linear regression). The structure of input data and error discovered through fuzzy clustering is captured in the form of a collection of fuzzy clusters, which helps eliminate (compensate) error produced by the global model. We discuss a detailed architecture of the proposed rule-based model and present its design based on an augmented version of Fuzzy C-Means (FCM). An extended suite of experimental studies offering some comparative analysis is covered as well.
•Developed is a concept of incremental models with fuzzy rules.•Rules capture the structure of error and realize its compensation.•Fuzzy clustering is used as a generic method to structure errors of the generic model.•A series of experimental studies is reported demonstrating the performance of the proposed approach.