Four-Valued Fuzzy Feature Fusion for Explainable Disease Classification: Multi-Dataset Evaluation and External Clinical Validation

High-dimensional healthcare datasets often contain correlated clinical variables that reduce model interpretability and increase computational complexity, limiting the development of trustworthy machine learning models for clinical decision support. To address this challenge, we propose a Four-Valued Fuzzy Feature Fusion (FVFFF) framework for explainable disease classification that transforms continuous clinical variables into four interpretable linguistic states, namely Absent, Mild, Moderate, and Severe, and aggregates clinically related attributes into compact fuzzy predictors. By integrating domain knowledge with fuzzy feature engineering, the proposed framework reduces feature dimensionality while preserving diagnostically relevant information for downstream machine learning models . The framework was evaluated using the Wisconsin Breast Cancer and Dermatology benchmark datasets together with an independent external clinical validation dataset collected from Fatima Medical Centre, Pakistan. Five machine learning algorithms, namely Random Forest, Support Vector Machine, XGBoost, LightGBM, and CatBoost, were evaluated using both conventional and fuzzy feature representations. Model interpretability was enhanced through SHAP-based explainability analysis, enabling the identification of clinically meaningful predictors that contribute to disease classification decisions. Experimental results demonstrate that the proposed FVFFF framework substantially reduces feature dimensionality while maintaining competitive predictive performance across multiple datasets. For the Wisconsin Breast Cancer dataset, the original 30 clinical variables were reduced to five interpretable fuzzy predictors while preserving strong classification performance. Similar improvements in computational efficiency and interpretability were observed for the Dermatology dataset and the independent external clinical validation dataset. SHAP analysis further confirmed that the generated fuzzy predictors retained clinically meaningful information,thereby supporting transparent and trustworthy healthcare decision-making.