Scholarlink

<p>Differentiation between Iron Deficiency Anemia (IDA) and Thalassemia trait remains a clinical challenge due to overlapping hematological profiles. This study proposes an interpretable machine learning framework that leverages ensemble learning and explainable artificial intelligence (XAI) for automated classification of IDA and Thalassemia using routine blood parameters. A dataset comprising 356 patient samples with 23 hematological features was analyzed, and multiple machine learning algorithms were systematically evaluated. Among the individual classifiers, tree‑based ensemble models such as Random Forest and CatBoost demonstrated strong performance, while a stacked ensemble integrating Random Forest, Logistic Regression, CatBoost, LightGBM, and K‑Nearest Neighbors achieved the best results with 92% accuracy and a ROC‑AUC of 0.96 on the held‑out test set. Explainability techniques including SHAP, LIME, and surrogate decision tree analysis were used to examine feature contributions and improve transparency of model behavior. The results indicate that interpretable machine learning models can effectively capture patterns within routine hematological parameters for distinguishing IDA from Thalassemia trait in the studied dataset. Further validation using larger and more diverse cohorts is required to assess generalizability and potential clinical applicability.</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

Interpretable ensemble learning for the classification of iron deficiency anemia and thalassemia via blood markers

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月21日

查看全文
  • Reva Sameer Laghate,
  • B. S. Dhruva Darshan,
  • Krishnaraj Chadaga,
  • Abhijay N. S,
  • Sushma Belurkar,
  • Niranjana Sampathila

摘要

Differentiation between Iron Deficiency Anemia (IDA) and Thalassemia trait remains a clinical challenge due to overlapping hematological profiles. This study proposes an interpretable machine learning framework that leverages ensemble learning and explainable artificial intelligence (XAI) for automated classification of IDA and Thalassemia using routine blood parameters. A dataset comprising 356 patient samples with 23 hematological features was analyzed, and multiple machine learning algorithms were systematically evaluated. Among the individual classifiers, tree‑based ensemble models such as Random Forest and CatBoost demonstrated strong performance, while a stacked ensemble integrating Random Forest, Logistic Regression, CatBoost, LightGBM, and K‑Nearest Neighbors achieved the best results with 92% accuracy and a ROC‑AUC of 0.96 on the held‑out test set. Explainability techniques including SHAP, LIME, and surrogate decision tree analysis were used to examine feature contributions and improve transparency of model behavior. The results indicate that interpretable machine learning models can effectively capture patterns within routine hematological parameters for distinguishing IDA from Thalassemia trait in the studied dataset. Further validation using larger and more diverse cohorts is required to assess generalizability and potential clinical applicability.