A Novel Fuzzy Concept-Cognitive Learning Model With Attribute Fluctuation and Concept Clustering

Concept-cognitive learning (CCL) is a paradigm that simulates human concept learning by processing given cues through specific cognitive models. However, existing CCL models face significant limitations, such as weak correlations between attributes and decisions, high redundancy within the concept space, and suboptimal learning performance. To address these issues, this article introduces an Attribute Fluctuation-Based CCL (AFFCCL) model. First, a novel measurement method for attribute fluctuation is proposed, based on the variation range of attribute membership degrees. To mitigate redundancy in the concept space, a fuzzy granular concept space is constructed using the concept contribution degree. Second, the model leverages the semantic richness of concepts by integrating similar fuzzy granular concepts, thereby constructing a clustering space. From this, upper and lower approximation spaces are derived. Finally, extensive experiments conducted on multiple benchmark datasets demonstrate that the proposed AFFCCL model outperforms representative fuzzy CCL models, neural network-based classifiers, and traditional similarity-based approaches in termsof accuracy, interpretability, and robustness.