Machine Learning Models for Passive Pre-symptomatic Detection of Covid-19 from Smart Wearable Data

Sarwar, Atifa

Etd

Machine Learning Models for Passive Pre-symptomatic Detection of Covid-19 from Smart Wearable Data

Öffentlich Deposited

Covid-19, a recently discovered Influenza Like Illness (ILI), is an infectious disease caused by Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2). The severe consequences of the Covid-19 pandemic highlighted the need for low-cost, passive assessment methods to detect infected patients early to restrain disease spread. This doctoral dissertation investigates Machine Learning models for detecting Covid-19 from disruptions in longitudinal physiological signs (such as heart rate and steps) collected passively from consumer-grade smart wearables, without relying on prior history or human-reported symptoms. The first research thrust of this dissertation focuses on passive Covid-19 detection through predictive modeling using traditional machine learning algorithms. It also proposes CovidRhythm, a deep Gated Recurrent Unit (GRU)-based Multi-Head Self-Attention (MHSA) model, to identify Covid-19 in the pre-symptomatic phase using biobehavioral rhythmic features. Given the significant individual differences in vital sign manifestations, which hinders the generalizability of AI models, the second research thrust introduces MetaCovid, a deep adaptation framework that leverages meta-learning to address the inter-subject differences with minimal data for pre-symptomatic infection detection. Since delayed diagnoses contributed significantly to the Covid-19 pandemic, early detection—identifying infections before symptom onset— is crucial for disease control. Lastly, the dissertation explores earliest possible Covid-19 infection detection through EarlyDetect, a Reinforcement Learning-based Early Time Series Classification method, and ECovGNN, a Graph Neural Network-based method that leverages intra- and inter-subject similarities for Covid-19 detection. This dissertation achieves breakthroughs in pre-symptomatic Covid-19 screening by demonstrating effective detection three days before symptom onset using heart rate and steps values over 72-hour non-overlapping windows. We believe that findings of this dissertation will pave the way for timely disease detection, clinical management, and improved public health responses to future infectious diseases.

Creator