Computer Science Department , MS Thesis Presentation , DInesh J Kodwai, " CovidRhythmTab: COVID-19 Infection Detection Framework from Wearable Sensor Data using Tabular Neural Networks to Analyze BioBehavioral Rhythms"
1:00 p.m. to 2:00 p.m.
Dinesh J Kodwani
MS student
WPI – Computer Science Department
Monday, April 20, 2026
Time: 1:00 p.m. – 2:00 p.m.
Location: Fuller Labs 141
Advisor: Prof. Emmanuel Agu
Reader : Prof. Dmitry Korkin
Abstract:
The SARS-Cov-2 virus caused the COVID-19 pandemic between 2020 and 2023. While the pandemic has ended, new infections continue to manifest. Early infection detection can inform timely interventions to curb pandemics. While PCR tests are accurate, their results are delayed and they require the patient to be present in-person, which presents a burden. Passive, convenient, early infection detection methods are desirable to limit disease spread. Human bodies follow natural, periodic clocks or biorhythms that govern various functions including activity, body temperature, heart rate and sleep-wake cycles.
Prior work has found that COVID-19 disrupted the natural body rhythms of infected patients, even before biological symptoms fully manifest. Many smartwatches, which are substantially cheaper than PCR machines, can continuously measure physiological signals such as the heart rate and activity of users. Such smartwatch data and extracted features are often recorded in a tabular data format that can be analyzed to detect diseases such as COVID-19.
The goal of this thesis is to investigate whether TabR, a Neural Network designed for tabular data can enhance the performance in pre-symptomatic prediction of COVID-19 from disruptions in biobehavioral rhythms of the users.
This thesis proposes CovidRhythmTab, a neural network framework that utilizes TabR to classify presymptomatic COVID-19 disease from smartwatch data. Rigorous evaluation was done including a comparison to state-of-the-art baselines. To the best of our knowledge this is the first work to explore the use of tabular neural networks to classify COVID-19 infection from smartwatch-based biobehavioral rhythmic features.