ThyroCare: An Ensemble Machine Learning Framework for AI-Driven Thyroid Disorder Screening with SHAP-Based Explainability

Thyroid disease is one of the most common endocrine disorders affecting millions worldwide, often leading to severe health complications if not diagnosed early. In this paper, we propose ThyroCare, a comprehensive machine learning-based framework for the efficient and accurate prediction of thyroid disorders using structured clinical data. Our approach integrates Principal Component Analysis (PCA) for dimensionality reduction, alongside an ensemble of four classifiers — Random Forest, Support Vector Machine (SVM), Decision Tree, and Logistic Regression — operating under a hard-voting majority consensus mechanism. Trained on a publicly available clinical dataset of 3,771 patient records with 29 features including TSH, T3, TT4, T4U, and FTI levels, the ensemble achieves an overall classification accuracy of 85.71% with a precision of 69.01% and specificity of 96.55% under controlled measurement perturbation. A SHAP-based explainability layer identifies the top contributing features per prediction, enabling transparent, patient-friendly clinical insights along with personalized dietary and consultation recommendations. An OCR-enabled medical report digitization module further enhances clinical utility by automating data extraction from uploaded laboratory reports. ThyroCare serves as a reliable, explainable, and deployable diagnostic support tool for early thyroid disease detection in real-world healthcare settings.