Machine Learning Models for Predicting Hypothyroidism: Utilizing Synthetic Data for Improved Accuracy

If left untreated, hypothyroidism, a common endocrine illness, can result in a number of health issues due to the underactive thyroid gland. Accurately predicting hypothyroidism is essential for prompt diagnosis and successful treatment of the illness. This work addresses class imbalance in hypothyroid datasets with and without SMOTE, examining the predictive power of the SVM, LR, RF, KNN and XgBoost algorithms. Models are assessed and trained using performance metrics that include F1-score, accuracy, precision, recall, and area under the ROC curve (AUC) using a dataset that includes clinical characteristics and thyroid function tests. According to the study's findings, all three models show excellent accuracy. Random Forest performs better than SVM and LR, especially when SMOTE is included, and provides better predictive capacities for the diagnosis of hypothyroidism.The work emphasizes the need of addressing class imbalance in medical predictive modeling tasks and offers useful information for the diagnosis and treatment of thyroid diseases.