Design and implementation of a hybrid machine learning framework for predicting heart rate status

Area of Science:

• Cardiovascular Health
• Machine Learning
• Physiological Monitoring

Background:

• Accurate real-time heart rate evaluation is crucial for physiological analysis, performance optimization, and early cardiovascular disease detection.
• Existing methods may face challenges in computational constraints and efficient deployment on embedded systems.

Purpose of the Study:

• To develop a lightweight hybrid machine learning model for classifying heart rate status (normal vs. warning) during physical activity.
• To improve the accuracy and efficiency of real-time heart rate monitoring, particularly for embedded platforms.

Main Methods:

• A weighted voting-based ensemble model combining Multilayer Perceptron, Naïve Bayes, and K-Nearest Neighbors algorithms.
• Analysis of physiological and environmental parameters including temperature, humidity, speed, incline, and activity time.
• Statistical analysis, including Shapiro-Wilk tests and hypothesis testing, to validate input variables.

Main Results:

• The ensemble classifier achieved up to 96.67% accuracy, 96.66% F1-score, and 0.9354 MCC in simulations.
• The embedded system achieved 90.83% accuracy and 0.8167 MCC, demonstrating real-time processing capabilities with slight performance degradation.
• The model outperforms standalone classifiers, indicating excellent classification balance and statistical significance.

Conclusions:

• The proposed hybrid machine learning model offers a robust solution for real-time heart rate monitoring.
• The framework effectively addresses computational constraints, enabling efficient deployment on embedded platforms like Arduino.
• This work contributes to advancements in wearable health technology and early cardiovascular risk assessment.