Scholarlink

<p>Thyroid disease detection involves clinical evaluation, lab tests, and imaging to identify thyroid disorders. Early detection is vital for timely treatment, preventing complications, and improving quality of life. However, current techniques face challenges such as non-specific symptoms, test inconsistency, limited access, and diagnostic limitations, affecting accuracy and timely diagnosis. In this research, a novel Deep High-order Attention Network (DHA-Net) optimized utilizing the Wolverine-Kite Optimization Algorithm (WKOA) is proposed to address these issues. The dataset, containing clinical data for thyroid disease detection, undergoes processing with tanh normalization to normalize the data. Feature selection is then performed with WKOA, a hybrid optimization devised using the Wolverine Optimization Algorithm (WoOA) as well as the Black Kite Algorithm (BKA). Next, data augmentation is performed using Local Density Estimation-based SMOTE (LD-SMOTE) to enhance the minority sample representation. The DHA-Net, with hyperparameters optimized via WKOA, is then used for thyroid disease detection. Model interpretability is ensured through Explainable AI (XAI) using SHapley Additive exPlanations (SHAP), providing insights into the contributions of each feature. The devised WKOA_DHA-Net model attains a True-Negative Rate (TNR) of 93.283%, True-Positive Rate (TPR) of 93.945%, overall accuracy of 93.435%, and F1-score of 93.387%.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

WKOA: Wolverine-Kite Optimization Algorithm for Feature Selection and Deep High-order Attention Network with Explainable AI for Thyroid Disease Detection

  • 期刊论文

  • 发表日期：

    2025年10月23日

查看全文
  • Mubin Tamboli,
  • Geeta S. Navale,
  • Priya Shelke,
  • Amol V. Dhumane

摘要

Thyroid disease detection involves clinical evaluation, lab tests, and imaging to identify thyroid disorders. Early detection is vital for timely treatment, preventing complications, and improving quality of life. However, current techniques face challenges such as non-specific symptoms, test inconsistency, limited access, and diagnostic limitations, affecting accuracy and timely diagnosis. In this research, a novel Deep High-order Attention Network (DHA-Net) optimized utilizing the Wolverine-Kite Optimization Algorithm (WKOA) is proposed to address these issues. The dataset, containing clinical data for thyroid disease detection, undergoes processing with tanh normalization to normalize the data. Feature selection is then performed with WKOA, a hybrid optimization devised using the Wolverine Optimization Algorithm (WoOA) as well as the Black Kite Algorithm (BKA). Next, data augmentation is performed using Local Density Estimation-based SMOTE (LD-SMOTE) to enhance the minority sample representation. The DHA-Net, with hyperparameters optimized via WKOA, is then used for thyroid disease detection. Model interpretability is ensured through Explainable AI (XAI) using SHapley Additive exPlanations (SHAP), providing insights into the contributions of each feature. The devised WKOA_DHA-Net model attains a True-Negative Rate (TNR) of 93.283%, True-Positive Rate (TPR) of 93.945%, overall accuracy of 93.435%, and F1-score of 93.387%.