Hybrid Ensemble Models for Early Risk Detection in Wearable Health Tech
A research collaboration by Sonny Marmon & Harshith Guduru
This project leverages hybrid ensemble machine learning models to predict early-stage health risks using real-world data from wearable devices. By fusing biosignals such as ECG, PPG, and accelerometry from the DREAMT Dataset (PhysioNet), we aim to build a scalable, interpretable framework for early health event detection and wellness insights.
We use the DREAMT Dataset, a publicly available multi-sensor dataset for sleep stage classification, which serves as a strong proxy for cardiovascular, metabolic, and neurological health indicators.
Features include:
- Electrocardiogram (ECG)
- Photoplethysmography (PPG)
- Accelerometry (x/y/z)
- Sleep stage labels
Access the dataset via PhysioNet
We develop and benchmark a hybrid ensemble learning pipeline using:
- Gradient Boosted Trees (XGBoost)
- Temporal Convolutional Networks (TCNs)
- LSTM-RNNs
- Soft Voting / Stacked Ensemble
Each model is evaluated using:
- Accuracy
- Precision/Recall
- F1-Score
- ROC-AUC
- Class-wise confusion matrix
- Detect early physiological deviations linked to potential risks
- Build a transparent and explainable ML system for wearable health insights
- Optimize for deployment on low-power edge devices (future work)
This project uses open-source, de-identified data only. No private medical data is accessed. All experiments comply with open-data licensing and research reproducibility principles.
-
Sonny Marmon
Student Researcher | AI, Quantum Systems, MS&E
LinkedIn • Website -
Harshith Guduru
Research Intern @ Walmart Global Tech | Published in Science under Nature
LinkedIn
This project is licensed under the MIT License – see the LICENSE file for details.
Marmon, S., & Guduru, H. (2025). Hybrid Ensemble Models for Early Risk Detection in Wearable Health Tech. GitHub. https://github.com/sonnymarmon/wearable-risk-ensemble