Self-learning interval type-2 hierarchical fuzzy system based on rule relevance with online regression prediction application

In this paper, a novel self-learning design of an interval type-2 hierarchical fuzzy system (IT2 HFS) based on rule relevance is proposed. Different from the existing methods, this paper uses data stream instead of batch data to construct the IT2 HFS. As time goes on, HFS gradually fills its rule base through fuzzy partition algorithm, where the rule base of HFS is empty at the beginning. A new self-updating method for consequent parameters is proposed based on rule relevance. When HFS is used for online learning, the proposed method can ensure that the system has better performance. The designed self-learning IT2 HFS is applied to online regression prediction issue, and is tested on two different online regression datasets. Self-learning type-1 HFS is also compared with self-learning IT2 HFS in prediction accuracy and system complexity. The results illustrate that the self-learning IT2 HFS performs better in prediction accuracy with lower rules.