<scp>CovidXAI</scp>: explainable<scp>AI</scp>assisted web application for<scp>COVID</scp>‐19 vaccine prioritization
Deepraj Chowdhury, Saranda Poddar, Soham Banarjee, Riya Pal, Abrar Gani, Caroline Ellis, Rajesh Arya, Sukhpal Singh Gill, Steve Uhlig
Abstract
COVID‐19 vaccines have a limited supply, and there is a huge gap between supply and demand, leading to disproportionate administration. One of the main conditions on which balanced and optimal vaccine distribution depends are the health conditions of the vaccine recipients. Vaccine administration of front‐line workers, the elderly, and those with diseases should be prioritized. To solve this problem, we proposed a novel architecture called CovidXAI, which is trained with a self‐collected dataset with 24 parameters influencing the risk group of the vaccine recipient. Then, Random Forest and XGBoost classifiers have been used to train the model—having training accuracies of 0.85 and 0.87 respectively, to predict the risk factor, classified as low, medium, and high risk. The optimal vaccine distribution can be done using the derived from the predicted risk class. A web application is developed as a user interface, and Explainable AI (XAI) has been used to demonstrate the varying dependence of the various factors used in the dataset, on the output by CovidXAI.