Leveraging machine learning and self-administered tests to predict COVID-19: An olfactory and gustatory dysfunction assessment through crowd-sourced data in India

It has been established that smell and taste loss are frequent symptoms during COVID-19 onset. Most evidence stems from medical exams or self-reports. The latter is particularly confounded by the common confusion of smell and taste. Here, we tested whether practical smelling and tasting with household items can be used to assess smell and taste loss. We conducted an online survey and asked participants to use common household items to perform a smell and taste test. We also acquired generic information on demographics, health issues including COVID-19 diagnosis, and current symptoms. We developed several machine learning models to predict COVID-19 diagnosis. We found that the random forest classifier consistently performed better than other models like support vector machines or logistic regression. The smell and taste perception of self-administered household items were statistically different for COVID-19 positive and negative participants. The most frequently selected items that also discriminated between COVID-19 positive and negative participants were clove, coriander seeds, and coffee for smell and salt, lemon juice, and chillies for taste. Our study shows that the results of smelling and tasting household items can be used to predict COVID-19 illness and highlight the potential of a simple home-test to help identify the infection and prevent the spread.