AdvancedHybridNet: An AI-Powered Hybrid Ensemble for High-Accuracy Thyroid Disease Diagnosis Using Dynamic Feature Selection

Abstract Thyroid diseases are increasingly prevalent, making early and accurate diagnosis critical to reducing mortality and complications. This study proposes Advanced-HybridNet, a novel machine learning framework that combines dynamic feature selection and an AI-driven ensemble learning approach to enhance thyroid disorder classification. By integrating DynamicRankSelector and a balanced class 1 sampling mechanism, the model effectively handles class imbalance and high-dimensional data, improving both robustness and interpretability. The ensemble comprises multiple optimized classifiers and leverages soft voting to enhance pre-dictive accuracy. The model achieves a sensitivity of 100% and an accuracy of 99.95%, outperforming conventional diagnostic methods. Additionally, random oversampling addresses class imbalance for minority conditions like hyperthy-roidism and hypothyroidism, leading to more stable predictions. Mathematical analysis further confirms the model’s high precision, reduced false positives, and consistent performance. These results suggest that AdvancedHybridNet offers a reliable, scalable, and more advanced diagnostic alternative to existing clinical techniques for thyroid disease prediction.