Scholarlink

<p>Predicting birth weight is a critical task in prenatal care, directly influencing maternal and neonatal health interventions and outcomes. While most prior studies have focused primarily on categorical classification of birth weight into discrete ranges (e.g., low, normal, high), the clinically more valuable task of predicting the exact birth weight through regression has received relatively little attention. Accurate regression-based prediction enables more precise risk stratification, individualized care planning, and timely intervention, which are essential for improving neonatal health. In this study, we address this gap by systematically evaluating a wide range of machine learning (ML), deep learning (DL), and ensemble regression models on a publicly available prenatal dataset. Among the models tested, a stacking ensemble combining Feedforward Neural Networks (FNN) and Gradient Boosting as base learners with LightGBM as the meta-model achieved the highest regression performance (R<sup>2</sup>: 0.5026, MSE: 0.8924), outperforming previously reported approaches. To ensure clinical interpretability, explainable AI (XAI) techniques, including SHAP, LIME, and Permutation Feature Importance, were incorporated to identify key prenatal factors contributing to birth weight, providing actionable insights for healthcare providers. Complementary experiments on classification into five clinically significant groups (using Random Forest, XGBoost, and LightGBM, achieving peak accuracy of 92%) and unsupervised clustering (K-Means, Bootstrap K-Means, DBSCAN, enhanced with PCA) demonstrated the robustness and versatility of the framework. This study emphasizes three major contributions: (i) highlighting regression-based birth weight prediction as a novel and underexplored approach, (ii) achieving superior predictive accuracy over existing methods, and (iii) integrating explainable AI to bridge predictive power with clinical interpretability. Overall, the proposed multi-task framework provides precise, interpretable, and clinically actionable predictions, representing a meaningful advancement in prenatal data analytics and maternal-neonatal healthcare.</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

A regression-based multi-task machine and deep learning framework for explainable AI driven birth weight prediction in maternal and neonatal care

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月22日

查看全文
  • Anika Bintee Aftab,
  • Nishat Anjum Lea,
  • Md. Rakibul Islam

摘要

Predicting birth weight is a critical task in prenatal care, directly influencing maternal and neonatal health interventions and outcomes. While most prior studies have focused primarily on categorical classification of birth weight into discrete ranges (e.g., low, normal, high), the clinically more valuable task of predicting the exact birth weight through regression has received relatively little attention. Accurate regression-based prediction enables more precise risk stratification, individualized care planning, and timely intervention, which are essential for improving neonatal health. In this study, we address this gap by systematically evaluating a wide range of machine learning (ML), deep learning (DL), and ensemble regression models on a publicly available prenatal dataset. Among the models tested, a stacking ensemble combining Feedforward Neural Networks (FNN) and Gradient Boosting as base learners with LightGBM as the meta-model achieved the highest regression performance (R2: 0.5026, MSE: 0.8924), outperforming previously reported approaches. To ensure clinical interpretability, explainable AI (XAI) techniques, including SHAP, LIME, and Permutation Feature Importance, were incorporated to identify key prenatal factors contributing to birth weight, providing actionable insights for healthcare providers. Complementary experiments on classification into five clinically significant groups (using Random Forest, XGBoost, and LightGBM, achieving peak accuracy of 92%) and unsupervised clustering (K-Means, Bootstrap K-Means, DBSCAN, enhanced with PCA) demonstrated the robustness and versatility of the framework. This study emphasizes three major contributions: (i) highlighting regression-based birth weight prediction as a novel and underexplored approach, (ii) achieving superior predictive accuracy over existing methods, and (iii) integrating explainable AI to bridge predictive power with clinical interpretability. Overall, the proposed multi-task framework provides precise, interpretable, and clinically actionable predictions, representing a meaningful advancement in prenatal data analytics and maternal-neonatal healthcare.