Stress Detection from Wearable Sensor Data (PPG + Accelerometer)

We are a health-tech startup exploring whethr wearable sensor data can reliably distinguish stress states from baseline in real-world conditions. Before committing to a full pipeline build, we want a proof-of-concept notebook that shows feasibility using a public dataset. Scope: Use an open-source wearable dataset (eg. WESAD or similar) containing PPG, accelerometer, and physiological signals with labeled stress/baseline segments Preprocess raw sensor streams: filtering, artifact rejection, segmentation into usable windows Extract time-domain and frequency-domain features from the physiological signals Train and evaluate at least one classification model for stress vs. baseline, using proper per-subject cross-validation Provide a brief analysis of which features contribute most to the classification Deliver one clean, documented Jupyter notebook that we can run and extend Required: Signal processing experience (filtering, spectral analysis, feature extraction from noisy sensor data) ML classification on time-series data (Python, scikit-learn, PyTorch or TensorFlow) Familiarity with proper evaluation methodology for subject-level physiological data Preferred: Background in electrical engineering or biomedical engineering Experience with wearable sensor data or physiological signals Familiarity with explainability methods