Predicting Health Risks with Integrated EHR and Wearable Data Using XGBoost and LSTM
- Version
- Download 10
- File Size 396.97 KB
- File Count 1
- Create Date 11 June 2025
- Last Updated 11 June 2025
Predicting Health Risks with Integrated EHR and Wearable Data Using XGBoost and LSTM
Authors:
Yalamakuri Suresh
Department of Artificial Intelligence and Data Science Central University of Andhra Pradesh, Ananthapuramu, India Email: suresh.yalamakuri@cuap.ac.in
Dr. P. Sumalatha
Department of Artificial Intelligence and Data Science Central University of Andhra Pradesh, Ananthapuramu, India Email: sumalatha.p@cuap.ac.in
Abstract—Heterogeneous data integration from electronic health records (EHRs) and wearable devices enhances predictive modeling in healthcare. This study proposes a hybrid model combining XGBoost and Long Short-Term Memory (LSTM) networks to predict health risks by leveraging structured EHR data (e.g., age, BMI, blood pressure) and time-series wearable data (e.g., heart rate, SpO2). The ensemble model averages prob- abilistic predictions from both models, achieving an AUC-ROC of 0.92, surpassing individual model performances (XGBoost: 0.85, LSTM: 0.88). Feature importance from XGBoost and temporal pattern analysis from LSTM provide interpretable insights, supporting clinical decision-making. This approach demonstrates the potential of multi-modal learning for personalized medicine and real-time risk stratification.
Index Terms—Electronic Health Records, Wearable Devices, XGBoost, LSTM, Health Risk Prediction, Multi-Modal Data Integration, Ensemble Model, Predictive Modeling