A Dynamic Selection Hybrid Model for Advancing Thyroid Care With BOO-ST Balancing Method

Abstract

Recently, thyroid disease has been a leading cause of mortality, underscoring the importance of early diagnosis to mitigate its impact. Researchers have randomly employed static selection ensemble methods aiming to forecast the disease in its initial stages. However, the use of such ensemble methods in healthcare diagnosis poses challenges related to performance consistency and potential mismatches with new data characteristics. Hence, this paper proposes a novel approach by introducing the Dynamic Selection Hybrid Model (DSHM) that leverages the most effective conventional classifiers using an appropriate ensemble method. Instead of going the conventional way, we evaluate various baseline classifiers to demonstrate their impact on the characteristics selected by two robust feature selection techniques. This evaluation employs an explainable AI (XAI) method, Permutation Feature Importance (PFI), and selects the most effective classifiers based on their characteristics impact. Then the selected classifiers are integrated by an appropriate ensemble method, based on a comparative evaluation between four efficient ensemble methods. Allowing the proposed DSHM to dynamically adjust its composition based on selecting conditions can potentially achieve robust performance by better adaptability on unseen data. Before the training DSHM, the methodology begins by addressing the dataset’s imbalance issue using an effective data balancing method BOO-ST. To demonstrate the superiority of DSHM, various performance evaluation matrices, and a statistical test are employed. The experimental results reveal the effectiveness of our proposed DSHM, outperformed with an impressive 99.33% accuracy. Finally, to enhance transparency, trust, and patient outcomes, we applied the Local Interpretable Model-agnostic Explanations (LIME) to explain DSHM-provided outcomes. With a robust classification performance, our proposed DSHM aims to explain its outcomes, contributing to improved clinical decision-making processes and ultimately enhancing patient care.

Keywords

• Dynamic selection hybrid classifier
• feature selection
• explainable AI
• thyroid disease