An Interpretable CatBoost Based Model for Accurate Classification of Thyroid Disorders

Thyroid diseases are among the most common endocrine disorders affecting a large portion of the global population. Conditions such as hypothyroidism and hyperthyroidism often remain undiagnosed because their symptoms overlap with other illnesses, making early detection challenging. To improve diagnostic accuracy, this study presents a CatBoost-based machine learning model for classifying thyroid conditions. A hybrid feature selection approach is applied to identify the most relevant clinical features and enhance model performance. To ensure interpretability, explainable AI (XAI) techniques, specifically LIME, are integrated to explain individual predictions and highlight the contribution of key features influencing the model’s output. Experimental evaluation demonstrates that the proposed model achieves a remarkable classification accuracy of 99%, effectively distinguishing between different thyroid conditions. The combination of high predictive accuracy and interpretability makes the proposed approach a reliable and transparent diagnostic support tool for clinicians assessing thyroid disease.