Exploring the synergy of media awareness and quarantine classes in SiSAQEIHR model for pandemic control: A Deep LSTM-RNN predictions

Abstract

Media awareness and higher treatment rates are crucial in pandemics to prevent disease spread, but the synergy between quarantine and awareness is often neglected. This investigation explores the impact of media awareness on infectious diseases. For this purpose, a novel eight compartmental $S_{i}SAQEIHR$ type mathematical model including eight individual classes, namely Immature Susceptible, Susceptible, Quarantine, Awareness, Exposed, Infective, Hospitality, and Recovered has been presented to depict disease dynamics. It incorporates a sigmoid type treatment rate, enhancing realism. This offers a fresh perspective on the study of infectious disease transmission. Mathematical analysis confirms the system’s positivity and boundedness, ensuring its theoretical stability analysis of the disease-free equilibrium point and applicability in predicting epidemic outcomes. Using the normalized forward sensitivity index, we have obtained sensitivity indices for factors associated with the basic reproduction number ($R_0$) and identified the essential parameters that significantly impact the modification in disease dynamics. Also, the influence of the parameters on $R_0$ is systematically analyzed and characterized through Global Sensitivity Analysis. Furthermore, this study underscores the importance of accurately predicting epidemic spread to enable timely interventions, focusing on the pivotal roles of Awareness, Hospitalization, and Exposed classes in dynamic epidemic model. A deep Long Short-Term Memory (LSTM) based Recurrent Neural Network (RNN) model is proposed for precise spread pattern prediction, leveraging advanced deep learning capabilities in time series analysis. Six hyperparameter topologies are analyzed to optimize prediction accuracy. Five statistical accuracy metrics reveal RMSE below 0.02 and $R$ and $R^2$ values exceeding 0.99 for all classes, validating satisfactory predictive performance. A real-life COVID-19 modeling study demonstrates the proposed LSTM-based model’s effectiveness in capturing complex dynamics, surpassing earlier traditional neural network methods. Furthermore, sensitivity analysis, varying the vaccinated proportion, confirms the robustness of the proposed model system in the realm of public health decision-making.