Scholarlink

Anemia persists as a major public health crisis in Bangladesh, disproportionately affecting women of reproductive age, children under five, and adolescents. The condition’s high prevalence underscores the urgent need for efficient screening methods to enable early intervention. This study addresses these challenges by developing an XAI-enhanced framework that leverages routine hematological parameters for accurate anemia detection. Using clinical records from Bangladeshi populations, we built a Stacking ensemble model combining four machine learning algorithms (Support Vector Machine, Decision Tree, Logistic Regression, and K-Nearest Neighbors) enhanced with interpretability. Our comprehensive statistical approach incorporated Z-score based outlier removal, Pearson correlation for feature selection, Random Forest-based importance ranking, and statistical validation through Z-test and T-tests—ensuring robust model development. The exploratory data analysis and SHAP explanation revealed ‘Hemoglobin levels’ as the most significant predictor (risk factor), aligning with clinical understanding of anemia pathophysiology. Our ensemble model demonstrated exceptional predictive capability, achieving 99.67% accuracy and 99.64% F1-score, substantially outperforming individual constituent algorithms. To facilitate clinical adoption, we designed an interactive real-time prediction interface enriched with interpretable model outputs, enabling healthcare practitioners to gain insights into the underlying rationale of each prediction. This dual focus on accuracy and interpretability represents a significant advancement over traditional black-box approaches in medical AI. The proposed framework serves as both a screening tool and decision support system (DSS). By automating and standardizing anemia assessment while maintaining clinical interpretability, this solution has potential to improve screening coverage, reduce diagnostic delays, and ultimately contribute to better anemia management outcomes across Bangladesh and similar demographic contexts.

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

Ensemble Learning-Based Framework for Anemia Screening Using Clinical Hematological Parameters

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Pankaj Bhowmik,
  • Mostofa Kamal Nasir

摘要

Anemia persists as a major public health crisis in Bangladesh, disproportionately affecting women of reproductive age, children under five, and adolescents. The condition’s high prevalence underscores the urgent need for efficient screening methods to enable early intervention. This study addresses these challenges by developing an XAI-enhanced framework that leverages routine hematological parameters for accurate anemia detection. Using clinical records from Bangladeshi populations, we built a Stacking ensemble model combining four machine learning algorithms (Support Vector Machine, Decision Tree, Logistic Regression, and K-Nearest Neighbors) enhanced with interpretability. Our comprehensive statistical approach incorporated Z-score based outlier removal, Pearson correlation for feature selection, Random Forest-based importance ranking, and statistical validation through Z-test and T-tests—ensuring robust model development. The exploratory data analysis and SHAP explanation revealed ‘Hemoglobin levels’ as the most significant predictor (risk factor), aligning with clinical understanding of anemia pathophysiology. Our ensemble model demonstrated exceptional predictive capability, achieving 99.67% accuracy and 99.64% F1-score, substantially outperforming individual constituent algorithms. To facilitate clinical adoption, we designed an interactive real-time prediction interface enriched with interpretable model outputs, enabling healthcare practitioners to gain insights into the underlying rationale of each prediction. This dual focus on accuracy and interpretability represents a significant advancement over traditional black-box approaches in medical AI. The proposed framework serves as both a screening tool and decision support system (DSS). By automating and standardizing anemia assessment while maintaining clinical interpretability, this solution has potential to improve screening coverage, reduce diagnostic delays, and ultimately contribute to better anemia management outcomes across Bangladesh and similar demographic contexts.