Comparative Analysis of Machine Learning Techniques for Heart Rate Prediction Employing Wearable Sensor Data

Area of Science:

• Cardiovascular physiology
• Machine learning applications
• Wearable sensor technology

Background:

• Wearable technology provides real-time cardiovascular insights crucial for health and athletic performance.
• Accurate heart rate (HR) monitoring is essential for personalized fitness and preventive healthcare strategies.

Purpose of the Study:

• To compare various machine learning (ML) techniques for HR prediction using wearable sensor data.
• To develop and evaluate a hybrid Singular Spectrum Analysis (SSA)-Augmented ML model for enhanced HR prediction.
• To assess the impact of incorporating auxiliary physiological data (breathing rate, RR intervals) on HR prediction accuracy.

Main Methods:

• Comparison of Long Short-Term Memory (LSTM) networks, Physics-Informed Neural Networks (PINNs), and 1D Convolutional Neural Networks (1D CNNs).
• Development of a hybrid SSA-LSTM model for HR prediction.
• Utilized cardiorespiratory data from 126 recordings of 81 participants across 10 sports, collected at 1 Hz.
• Investigated the effect of including breathing rate (BR) and RR intervals as auxiliary inputs.

Main Results:

• The hybrid SSA-LSTM model demonstrated the lowest prediction error, effectively capturing complex HR dynamics.
• Integrating HR, BR, and RR data significantly improved prediction accuracy compared to single or dual parameter inputs.
• The model showed effectiveness across participants with varying sports experience levels.

Conclusions:

• Multivariate machine learning models, particularly the hybrid SSA-LSTM approach, are highly effective for accurate HR prediction.
• Incorporating multiple physiological parameters enhances the reliability of wearable-based health monitoring systems.
• These findings support the adoption of advanced ML for improved fitness tracking and preventive healthcare.