Personalized Heart Rate Variability and Machine Learning Framework for Monitoring Exercise-Induced Cardiovascular Strain

Overview

Physical activity guidelines don’t have clear thresholds for people at higher cardiac risk, including patients with spontaneous coronary artery dissection (SCAD). Patients are told to exercise "not vigorously" or “moderately” but aren’t given any concrete metrics. This project introduces a personalized method to translate heart rate variability metrics into practical assessment for physical activity tracking using data-driven analysis. A custom Android app was developed using the official Polar SDK to stream live output from a chest-worn ECG wearable (Polar H10). To validate the framework, publicly available physiological signals from the PhysioNet Pulse Transit Time PPG Dataset (Mehrgardt et al., 2022) were analyzed across rest, walking, and running activities for 22 subjects. Results showed that heart rate variability dropped substantially during running (mean reduction: 39%, Cohen's d = 0.90) with a graded response across activity intensities. A modality ablation confirmed that combining cardiac and accelerometer features (99.0%) outperforms either signal alone (74.2% cardiac-only, 97.4% accelerometer-only). A gradient-boosted classifier trained on 25 engineered features achieved 95.6% accuracy under leave-one-subject-out cross-validation across all 22 subjects. The decoupling between heart rate elevation and HRV suppression accounted for 90.3% of predictive importance. This shows that personalized feature engineering and cross-signal interactions give clearer insight into autonomic load than raw signals or any single modality alone.

Competition history

• CSEF 2026 Computational Science (Senior Division) · Entry S-07-39

Related projects

Closest projects by meaning, across every fair and year in the corpus.

Browse more like this

• All Computational Sciences projects
• CSEF Computational Sciences projects
• Every CSEF 2026 project