Enhancing Thyroid Disease Prediction: Integrating Machine Learning Models for Improved Diagnostic Accuracy and Interpretability
Keywords:
Thyroid Disease Prediction, Machine Learning, XGBoost, Random Forest, Gradient Boosting, Logistic Regression, SMOTE, SHAP, Explainable AI, Feature Engineering, Ensemble Learning, UCI Repository, Clinical Decision Support, Concept Drift, InterpretabilityAbstract
Thyroid disorders represent a substantial global public health burden, frequently remaining undetected until significant
clinical complications arise. This paper presents a comprehensive machine learning framework for early and accurate
thyroid disease prediction, systematically integrating four established classification algorithms: Logistic Regression,
Random Forest, Gradient Boosting, and XGBoost. The proposed system addresses the critical challenge of class imbalance
through the Synthetic Minority Over-sampling Technique (SMOTE), and employs rigorous preprocessing including missing
value imputation, categorical encoding, and recursive feature elimination. Feature interpretability is operationalized via
SHAP (SHapley Additive exPlanations) values, enabling clinical transparency and physician trust. Experimental evaluation
on the UCI Machine Learning Repository thyroid dataset demonstrates that the ensemble-based XGBoost model achieves
98.6% accuracy, 97.8% precision, 98.1% recall, and a macro F1-score of 97.9%, outperforming all individual classifiers
evaluated. The framework further incorporates a concept drift monitoring module to maintain adaptive model performance
in dynamic clinical environments. This integrated solution provides a robust, interpretable, and clinically deployable tool for
thyroid disease screening and diagnosis support.
