Scholarlink

The problem of delayed thyroid illness identification in spite of improvements in diagnostic methods is addressed in this study. To produce a dependable and understandable prediction model, it combines ensemble learning with a sophisticated feature engineering process. SMOTE reduces class imbalance and applies min-max scaling to input characteristics to standardize them, guaranteeing fair representation in medical datasets. The training subspace is refined using a two-phase feature selection method that uses Recursive Feature Elimination (RFE) and SelectKBest. The suggested hybrid ensemble framework, which combines Random Forest and Decision Tree (RFT+DT), achieves a peak classification accuracy of 99.65% when compared to five state-of-the-art machine learning classifiers. While L2 regularization manages overfitting, RandomizedSearchCV's hyperparameter adjustment maximizes model performance. Model transparency is improved by integrating SHAP and LIME, which promotes confidence in clinical applications. Extensive analyses based on ROC-AUC, F1-score, recall, precision, and accuracy show excellent discriminative power and flexibility, making this framework a game-changing tool for better patient care and early thyroid illness prediction.

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

Early Thyroid Disease Diagnosis Using a Hybrid Ensemble Learning Approach with Feature Selection, SMOTE, and Model Explainability

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Md. Tofael Ahmed Bhuiyan,
  • Shahriar Manzoor,
  • Nur AAlam Munna,
  • Khandaker Mohammad Mohi Uddin

摘要

The problem of delayed thyroid illness identification in spite of improvements in diagnostic methods is addressed in this study. To produce a dependable and understandable prediction model, it combines ensemble learning with a sophisticated feature engineering process. SMOTE reduces class imbalance and applies min-max scaling to input characteristics to standardize them, guaranteeing fair representation in medical datasets. The training subspace is refined using a two-phase feature selection method that uses Recursive Feature Elimination (RFE) and SelectKBest. The suggested hybrid ensemble framework, which combines Random Forest and Decision Tree (RFT+DT), achieves a peak classification accuracy of 99.65% when compared to five state-of-the-art machine learning classifiers. While L2 regularization manages overfitting, RandomizedSearchCV's hyperparameter adjustment maximizes model performance. Model transparency is improved by integrating SHAP and LIME, which promotes confidence in clinical applications. Extensive analyses based on ROC-AUC, F1-score, recall, precision, and accuracy show excellent discriminative power and flexibility, making this framework a game-changing tool for better patient care and early thyroid illness prediction.