A data science and machine learning research project focused on non-invasive human mental stress identification using multi-modal physiological sensor streams collected from consumer wearables.
Key Highlights
- Sensor Modalities: Analyzed wearable time-series datasets featuring Photoplethysmography (PPG) heart rate, Heart Rate Variability (HRV), Galvanic Skin Response (GSR), skin temperature, and 3-axis accelerometer movement.
- Signal Processing & Feature Extraction: Extracted statistical time-domain and frequency-domain biometric indicators (RMSSD, SDNN, spectral power) to identify autonomic nervous system stress reactions.
- Classification Models: Compared tree-based classifiers (Random Forest, XGBoost) and neural architectures for continuous stress vs. baseline state classification.
- Preventative Health Potential: Investigated the viability of deploying lightweight quantized models directly on edge devices for real-time wellness intervention prompts.