Recursive Stacked LeakyNet-Based Compartmental Model for Analysis of COVID-19 Transmission Dynamics

Abstract

Mathematical models involving complex formulations are used for solving estimation problems across various disciplines. Epidemiological models like susceptible–infected–removed (SIR), and its variations like susceptible–exposed–infected–removed (SEIR) and others are traditionally used for estimating the progression of viral outbreaks. But, real-life situations adapt and respond to ever-changing surroundings and social norms in precise or random or in complex manners. The mathematical formulation-based compartmental models suffer from drawbacks that lead to inaccuracy in predictions. To address this gap, a deep learning (DL)-based hybrid architecture namely residual leaky network (LeakyNet) has been proposed for the first time to the best of authors’ knowledge. This DL-based hybrid architecture is capable of analysing various aspects of the pandemic by training on short-term data to generate probable future trends. Using this hybrid architecture as a building block, a DL-based epidemiological model namely vulnerable–contaminated–immuned (VCI) model has been formulated to analyse short-term historical data of COVID-19. The DL empowered compartmental model has been implemented on an ongoing real-life scenario and the results reveal that the proposed model performs reasonably well in comparison to the contemporary models. The predicted data have been observed to be in good agreement with the actual pandemic data. The prediction models considered in this study have been evaluated using mean absolute percentage error (MAPE).