Early prediction of hypothyroidism based on feature selection and explainable artificial intelligence

Abstract

Early and accurate diagnosis is required for adequate treatment of hypothyroidism. However, the presence of subjectivity in the interpretation of test results presents a significant challenge. In this study, we explored and evaluated the potential of machine learning (ML) algorithms for addressing this issue. These algorithms include decision trees, random forest, XGBoost, LightGBM, extra trees, gradient boosting, and a stacking ensemble model. The purpose is to predict hypothyroidism, which is a medical condition that affects the thyroid gland, using attributes derived from blood test results. These attributes include thyroxine, thyroid stimulating hormone, free thyroxine index, total thyroxine, and triiodothyronine. The results demonstrate the effectiveness of utilizing these algorithms for accurately classifying hypothyroidism and offering diagnostic assistance with 99.16% of accuracy.