Publication

Abstract : Hypothyroidism is a common endocrine disease that often presents with nonspecific symptoms, making timely and accurate diagnosis difficult. Conventional approaches to diagnosis are costly, and often times invasive. Therefore, reliable computational methods need to be developed. In this paper, an end-to-end machine learning pipeline is proposed for improved prediction of hypothyroidism while addressing the specific challenges of high-dimensional clinical feature sets and class imbalance. The training set is resampled using the Synthetic Minority Oversampling Technique (SMOTE), and feature selection is accomplished using feature analysis based on the Analysis of Variance (ANOVA) F-test to identify the clinical biomarkers with the most information. Three supervised learning models: Logistic Regression, Support Vector Machine (SVM), and Random Forest, are trained and compared to one other. The results of the experiments revealed that the Random Forest classifier outperformed all other models by achieving a 99.0% accuracy and F1-score of 0.92. This study provides support for a feature engineered approach that can greatly improve diagnostic sensitivity, while establishing a reliable and interpretable framework for designing clinical decision support systems for thyroid disease.