Scholarlink

Diabetes mellitus is becoming an increasingly critical health issue across globe that demands early diagnosis and continuous monitoring. Although traditional machine learning models have been employed to predict diabetes, they frequently lack clarity and have difficulty in identifying uncommon or rare patient cases. This research proposes a novel explainable hybrid framework that combines deep learning-based temporal modeling with classical machine learning and unsupervised anomaly detection. The framework utilizes multimodal data sources, including static clinical features, time-series Electronic Health Records (HER) and wearable sensor data, for robust diabetic risk assessment. Explainability is achieved using SHAP (SHapley Additive exPlanations) and counterfactual reasoning to provide both global and local interpretability. In addition, an autoencoder-based novelty detection module identifies patients whose health patterns deviate significantly from the normal. Experimental results on benchmark datasets demonstrate improved prediction accuracy, better anomaly identification and enhanced interpretability making the model suitable for real-world clinical decision support systems.

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 Novel Explainable Hybrid Framework for Temporal Risk Prediction and Anomaly Detection in Diabetic Disease Using Multimodal Health Data

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • K. Shailaja,
  • Shirina Samreen

摘要

Diabetes mellitus is becoming an increasingly critical health issue across globe that demands early diagnosis and continuous monitoring. Although traditional machine learning models have been employed to predict diabetes, they frequently lack clarity and have difficulty in identifying uncommon or rare patient cases. This research proposes a novel explainable hybrid framework that combines deep learning-based temporal modeling with classical machine learning and unsupervised anomaly detection. The framework utilizes multimodal data sources, including static clinical features, time-series Electronic Health Records (HER) and wearable sensor data, for robust diabetic risk assessment. Explainability is achieved using SHAP (SHapley Additive exPlanations) and counterfactual reasoning to provide both global and local interpretability. In addition, an autoencoder-based novelty detection module identifies patients whose health patterns deviate significantly from the normal. Experimental results on benchmark datasets demonstrate improved prediction accuracy, better anomaly identification and enhanced interpretability making the model suitable for real-world clinical decision support systems.