Diabetes Classification Using SMOTE-Based Intuitionistic Fuzzy K-NN with Weighted HIAD

Abstract

Diabetes risk classification is challenged by class imbalance and diagnostic uncertainty. This study investigates an Intuitionistic Fuzzy Set framework integrated with a proposed Hausdorff-Inspired Attribute Distance within the K-Nearest Neighbor algorithm. The framework models patient profiles using membership, non-membership, and hesitation components to capture clinical ambiguity. We evaluated this approach on the Pima Indians Diabetes dataset using Stratified 5-Fold Cross-Validation, incorporating class balancing and clinically informed feature weighting. Results showed that while a conventional Euclidean-based model achieved the highest accuracy (73.96%) and data balancing maximized sensitivity (77.61%), the intuitionistic fuzzy configurations achieved the highest specificity (89.80%), indicating a conservative uncertainty-aware classification behavior. The proposed distance measure yielded classification performance highly comparable to that of the conventional Hamming distance, providing an alternative maximum-deviation-based similarity formulation within the intuitionistic fuzzy framework. Furthermore, combining class balancing with feature weighting contributed to a more balanced sensitivity–specificity profile. Overall, the findings demonstrate how uncertainty representation, feature weighting, class balancing, and neighborhood configuration influence classification outcomes, highlighting the potential of intuitionistic fuzzy approaches for uncertainty-aware medical decision-support systems.