Scholarlink

This Paper examines the integration of artificial Electronic Health Records (EHR) and hierarchical machine learning systems to promote the use of synthetic data in early disease detection. To address the ongoing lack of strong real-world medical data, we utilize generative adversarial networks (GANs), specifically the conditional texture GAN (CTGAN) GAN, to generate clinically plausible EHR datasets. This synthesized information was mixed with real-life clinical data to teach a sequence of base-level learning models. Meta-learning element then combines the predictions of these lower-level models, thus, creating a stacked ensemble that increases generalization and interpretability due to its variety. Real-life testing of the UCI Heart Disease observational dataset indicates that this framework significantly enhances diagnostic precision (maximum 81.97%) and area under the receiver operating curve (AUC; maximum 0.8885) and can handle a lack of data at the same time. Together, these findings support the ability of synthetic data to complement current datasets and render stacking methodologies useful in developing highly precise and trustworthy diagnostic AI models. This study, therefore, traces a possible course for more accurate and data-protecting analytics in healthcare.

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

Synthetic Patient Simulation and Model Stacking for Early Disease Detection: an EHR-Focused AI Framework for Diagnostic Accuracy and Generalization

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sathish Kumar Natarajan,
  • Azween Abdullah,
  • Prabhu Natarajan

摘要

This Paper examines the integration of artificial Electronic Health Records (EHR) and hierarchical machine learning systems to promote the use of synthetic data in early disease detection. To address the ongoing lack of strong real-world medical data, we utilize generative adversarial networks (GANs), specifically the conditional texture GAN (CTGAN) GAN, to generate clinically plausible EHR datasets. This synthesized information was mixed with real-life clinical data to teach a sequence of base-level learning models. Meta-learning element then combines the predictions of these lower-level models, thus, creating a stacked ensemble that increases generalization and interpretability due to its variety. Real-life testing of the UCI Heart Disease observational dataset indicates that this framework significantly enhances diagnostic precision (maximum 81.97%) and area under the receiver operating curve (AUC; maximum 0.8885) and can handle a lack of data at the same time. Together, these findings support the ability of synthetic data to complement current datasets and render stacking methodologies useful in developing highly precise and trustworthy diagnostic AI models. This study, therefore, traces a possible course for more accurate and data-protecting analytics in healthcare.