<![CDATA[Clinical variant interpretation is essential for precision medicine; however, conventional machine learning approaches often focus on prediction accuracy without providing sufficient interpretability and decision-support capabilities. This study proposes a hybrid framework integrating Light Gradient Boosting Machine (LightGBM), SHapley Additive exPlanations (SHAP), and a Fuzzy Decision Support System (FDSS) for clinical variant risk assessment using the ClinVar dataset. A stratified sample of 200,000 genetic variants was utilized for model development and evaluation. LightGBM was employed to predict variant pathogenicity, while SHAP was applied to identify feature contributions and improve model transparency. The resulting prediction probabilities were subsequently processed through fuzzy inference to generate interpretable risk categories and recommendation-oriented outputs. Experimental results showed that the proposed framework achieved an Accuracy of 95.89%, Precision of 95.58%, Recall of 82.97%, F1-Score of 88.83%, and ROC-AUC of 98.73%. Explainability analysis revealed that variant-type representation was the most influential predictor of pathogenicity. The proposed framework extends conventional classification by transforming predictive outputs into actionable risk assessments, thereby enhancing transparency and supporting informed genomic decision-making. These findings demonstrate the potential of integrating predictive analytics, explainable artificial intelligence, and fuzzy reasoning for clinical variant assessment in precision medicine.]]>
Copyrights © 2026
