<p>Thyroid disease (TD) is a prevalent health problem that requires accurate and quick diagnosis for effective treatment. In this paper, we explore machine learning (ML) models using feature selection (FS) techniques and the hybrid Particle Swarm and Grey Wolf Optimization (PSO-GWO) to enhance the classification accuracy of TD. Two publicly available thyroid datasets were employed: the Thyroid Disease Dataset from the Kaggle Repository and the Thyroid Dataset from the Figshare Repository. Binary Ant Colony Optimization (BACO), binary particle swarm optimization (BPSO), and Binary Cuckoo Search (BCS) were used to evaluate the best FS algorithm. The BACO achieved the lowest average error of 0.9186. Five ML models, Stochastic Gradient Descent Classifier (SGD), Support Vector Machine (SVM) Classifier, K-Nearest Neighbors (KNN), Extra Trees (ET) Classifier, and Logistic Regression (LR), were evaluated before and after FS. The experiment results showed significant performance improvements after FS. The first dataset, the ET, achieved the best accuracy of 0.9894. For the second dataset, the ET achieved near-perfect results with 0.9991 accuracy, 0.9991 sensitivity, 0.9989 specificity, and an AUC of 0.999998. After optimizing ET with PSO-GWO, PSO-GWO-ET achieves 99.99% accuracy on both datasets. The BACO with PSO-GWO improved the classification metrics. This paper used evaluation metrics such as accuracy, F1 score, sensitivity (recall), specificity, and AUC. These results demonstrate that FS with BACO can significantly improve the performance of ML models for thyroid disease prediction. The ET and KNN emerged as the best-performing models across both datasets, demonstrating their effectiveness in accurately classifying TD cases.</p>