Machine learning-based optimization of pre-symptomatic COVID-19 detection through smartwatch.

Area of Science:

• Biomedical Engineering
• Data Science
• Infectious Disease Surveillance

Background:

• Asymptomatic and pre-symptomatic individuals significantly contribute to COVID-19 transmission, especially following viral mutations like Delta.
• Current diagnostic methods often rely on symptom manifestation, delaying detection and intervention.
• Physiological data from smartwatches show potential for early anomaly detection indicative of infection.

Purpose of the Study:

• To propose and evaluate the One Class-Support Vector Machine (OC-SVM) algorithm for pre-symptomatic COVID-19 detection using smartwatch data.
• To compare the performance of OC-SVM against existing methods, such as the Mahalanobis distance-based approach.
• To identify optimal parameters for OC-SVM, specifically focusing on Resting Heart Rate (RHR) and moving average window sizes.

Main Methods:

• Application of the One Class-Support Vector Machine (OC-SVM) algorithm to physiological data streams from smartwatches.
• Utilizing Resting Heart Rate (RHR) as a key physiological indicator.
• Employing moving average window sizes of 350 and 300 for data smoothing and feature extraction.

Main Results:

• OC-SVM demonstrated superior performance compared to the Mahalanobis distance method in detecting COVID-19 infection.
• The proposed OC-SVM method achieved earlier detection (23.5-40% earlier) and improved detection rates (13.2-19.1% relative improvement).
• OC-SVM resulted in a reduction in false positive rates, enhancing diagnostic accuracy.

Conclusions:

• One Class-Support Vector Machine (OC-SVM) using Resting Heart Rate (RHR) with 350 and 300 moving average window sizes is a highly effective technique for pre-symptomatic COVID-19 detection.
• Smartwatch-derived physiological data, when analyzed with advanced machine learning, offers a promising avenue for proactive public health monitoring.
• The findings support the integration of wearable technology and AI for early identification of infectious diseases, mitigating spread.