A Framework for Predicting Pandemic Severity and Mortality Using Generative AI and LLMs

Abstract This research work introduces a comprehensive framework for pandemic severity and mortality prediction. The study achieves two key objectives. Firstly, a robust model is developed for accurate mortality prediction using deep learning mechanisms, leveraging variants like Convolutional Neural Networks, Recurrent Neural Networks, and Transformers which learn complex relationships from clinical and demographic data, enabling timely risk stratification. Secondly, enhancing model explainability through large language models (LLMs) for explainability. Integrating LLMs provides natural language explanations alongside predictions. Transparency is enhanced by describing factors influencing severity and mortality, bolstered by confidence scores and uncertainty estimates. The methodology involves data preprocessing, fine-tuning LLMs, model development using deep learning algorithms, evaluation of models using performance metrics, and qualitative analysis of LLM results via user surveys. The generative AI model emerges as a valuable tool for pandemic management, adaptable beyond COVID-19. The novelty of this study lies in three aspects. Firstly, proposing and implementing a pandemic severity and mortality prediction framework using neural networks and deep learning. Secondly, introducing natural language explanations using LLMs alongside predictions, enhancing transparency, and thirdly demonstrating the effectiveness of the generative AI model across various pandemic-causing viruses, facilitated by a generic and adaptable data schema and model architecture.