Neutrosophic Cosine Similarity Fusion with CRITIC-Weighted
Ideal Profile Matching for Multi-Attribute Diabetes Risk
Stratification: Evidence from the CDC BRFSS 2021 Dataset
Dae Yu Kim 1,∗, Jeong Chan Park2
1Department of Electrical Engineering, Inha University, Korea
2Central Asian University, Tashkent, Uzbekistan
Emails: dyukim@inha.ac.kr; goodnews1979@gmail.com
Abstract
Accurate stratification of diabetes risk requires integrating clinically heterogeneous indicators under conditions of measurement
ambiguity, borderline readings, and inconsistent self-reported data. This paper introduces a Neutrosophic Cosinesimilarity
with CRITIC-weighted ideal-profile matching (NCRS-CRITIC) framework that maps each patient record to an
ideal disease profile and an ideal healthy profile simultaneously, using neutrosophic truth, indeterminacy, and falsity membership
functions. The degree of closeness to each profile is measured through a weighted neutrosophic cosine similarity,
where feature weights are derived via the CRITIC (CRIteria Importance Through Intercriteria Correlation) method—
capturing both the discriminative variability and the inter-feature correlation structure objectively. A relative closeness
coefficient (RC) aggregates dual-profile similarity into a scalar risk score that respects both the evidence for and against
disease simultaneously. Experiments on a balanced 2000-instance subset of the CDC Behavioral Risk Factor Surveillance
System (BRFSS) 2021 Diabetes Health Indicators Dataset achieve an area under the ROC curve (AUC) of 0.869 and accuracy
of 79.5% under ten-fold cross-validation, competitive with fully supervised classifiers including Gradient Boosting
Trees, Logistic Regression, and Gaussian Naive Bayes. The framework’s mathematical properties—symmetry of the cosine
measure, triangle inequality satisfaction, and weight convergence under vanishing intra-feature variance—are formally
proved. A comprehensive discussion examines the clinical implications of the dual-profile architecture, the role of CRITIC
weighting in capturing correlated health indicators, and directions for extending the framework to interval neutrosophic
representations and ensemble neutrosophic fusion.
Keywords: Neutrosophic sets; Cosine similarity; information fusion; CRITIC weighting; Ideal solution; Diabetes prediction;
CDC BRFSS; Mlti-attribute decision making; Uncertainty modelling; Pattern recognition