Kutlu, FatihGoleli, KubraCastillo, Oscar2025-10-302025-10-3020251863-17031863-171110.1007/s11760-025-04784-32-s2.0-105016729550https://doi.org/10.1007/s11760-025-04784-3https://hdl.handle.net/20.500.14720/28785This study proposes a dual-stage optimization framework for uncertainty-aware classification by integrating the Intuition-istic Fuzzy Any Relation Clustering Algorithm (IF-ARCA) with Intuitionistic Fuzzy K-Nearest Neighbors (IF-KNN). In the first stage, Harris Hawks Optimization (HHO) calibrates IF-ARCA parameters to construct reliable membership and non-membership matrices, while in the second stage HHO independently tunes IF-KNN parameters, ensuring decoupled and stable convergence. HHO was chosen for its effective exploration-exploitation balance in high-dimensional search spaces, and the dual-stage design uniquely enables clustering and classification to be optimized without mutual interfer-ence. Extensive experiments on eight benchmark datasets (seven from UCI, plus Yeast and Credit Fraud for scalability) confirm the superiority of the proposed approach: the fuzzy metric variant achieved F1 = 0.993 on Credit Fraud and 0.946 on MONK's Problems, while cosine similarity reached 0.989 on Digits. Compared with established FKNN variants, the framework yielded 20-35% relative improvements and demonstrated statistically significant gains on challenging datas-ets (Iris, MONK's, Yeast; Wilcoxon p < 0.05). These results highlight the framework's robustness under class overlap and imbalance, while maintaining competitive performance in high-dimensional domains, establishing a novel contribution to clustering-guided classification and nature-inspired optimization.eninfo:eu-repo/semantics/closedAccessIntuitionistic Fuzzy SetsHarris Hawks OptimizationHybrid ClassificationFuzzy Metric SimilarityMachine Learning OptimizationArticle1913Q3Q2WOS:001576018300009