Scholarlink

<p>The increasing complexity of healthcare data demands predictive models that are both accurate and interpretable. This study presents the Clinical Predictive Fusion Network (CPFN). This adaptive ensemble learning framework integrates Logistic Regression, Random Forest, and Support Vector Machine classifiers through a validation-driven weighted fusion strategy. The model’s adaptive weighting enables it to learn the relative reliability of base classifiers across multimodal patient datasets. CPFN was evaluated using 10-fold stratified cross-validation on disease-specific (cardiology, neurology, diabetes, pulmonology, and oncology) and a synthetically fused multi-disease dataset, achieving up to 93.0 ± 0.4% accuracy on individual datasets and 95.5 ± 0.3% on the combined dataset. Other metrics included a recall of 92.0 ± 0.5%, F1-score of 92.5 ± 0.4%, and ROC-AUC ranging from 0.95 to 0.975 (95% CI, bootstrap 1000 resamples). These results demonstrate that CPFN maintains consistent and generalizable performance across heterogeneous data sources. The model’s transparent fusion design and detailed pseudocode enhance reproducibility and clinical applicability, positioning CPFN as a scalable, data-driven decision-support framework for next-generation predictive healthcare systems.</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

Clinical predictive fusion network for accurate disease prediction in patient cohorts

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月25日

查看全文
  • Md Zeyauddin,
  • Shafiqul Abidin,
  • Imran Khan,
  • Mohammad Ubaidullah Bokhari,
  • Md Ashraf Siddiqui,
  • Ausaf Ahmad,
  • Shadab Alam

摘要

The increasing complexity of healthcare data demands predictive models that are both accurate and interpretable. This study presents the Clinical Predictive Fusion Network (CPFN). This adaptive ensemble learning framework integrates Logistic Regression, Random Forest, and Support Vector Machine classifiers through a validation-driven weighted fusion strategy. The model’s adaptive weighting enables it to learn the relative reliability of base classifiers across multimodal patient datasets. CPFN was evaluated using 10-fold stratified cross-validation on disease-specific (cardiology, neurology, diabetes, pulmonology, and oncology) and a synthetically fused multi-disease dataset, achieving up to 93.0 ± 0.4% accuracy on individual datasets and 95.5 ± 0.3% on the combined dataset. Other metrics included a recall of 92.0 ± 0.5%, F1-score of 92.5 ± 0.4%, and ROC-AUC ranging from 0.95 to 0.975 (95% CI, bootstrap 1000 resamples). These results demonstrate that CPFN maintains consistent and generalizable performance across heterogeneous data sources. The model’s transparent fusion design and detailed pseudocode enhance reproducibility and clinical applicability, positioning CPFN as a scalable, data-driven decision-support framework for next-generation predictive healthcare systems.