Automated COVID-19 Detection From Exhaled Human Breath Using CNN-CatBoost Ensemble Model

In this letter, we propose a novel approach for detecting COVID-19 from exhaled breath. Recent studies have revealed that certain exhaled breath components can be used as biomarkers for detecting COVID-19. Breath-based analysis stands out among the various noninvasive ways of detection as it provides a simpler and more relaxed method of diagnosis. Low-cost diagnosis methods are required to prevent the spread of infectious diseases, such as COVID-19. A cost-effective sensing module consisting of an array of sensors is developed in this letter to measure the levels of acetone, methanol, isopropanol, and ethanol in the exhaled breath to detect COVID-19. For the best predictive performance, we have implemented a convolution neural network-categorical boosting (CNN-CatBoost) ensemble model, which classifies the test samples to make predictions. The proposed procedure correctly classified the samples with an accuracy of 96.15%. The results of the study suggest that the combinations of these four biomarkers, along with the proposed sensing module and deep learning technique, can serve as a trustworthy method for noninvasively detecting COVID-19.