Integrating XGBoost and Recurrent Neural Networks (RNNs) to Optimize COVID-19 Mortality Prediction

Abstract

The SARS-CoV-2 pandemic has severely affected worldwide health, but the prediction of patient mortality remains difficult to achieve accurately. Current predictive methods mainly concentrate on clinical condition measurements without a complete view of death predictions. This research fills the void between advanced modelling through XGBoost and Recurrent Neural Networks (RNNs) for improved COVID-19 mortality prediction analysis. We applied a systematic four-phase approach to collect data from the "COVID-19 Symptoms Dataset containing symptoms and death cases as well as mortality rates and confirmed cases, " prior to data processing and analysis stages, including cleaning, visualization and feature analysis, followed by the implementation and optimization of XGBoost models and RNN frameworks. The final stage involved performance assessment with accuracy, precision recall, and F1-score. The evaluation results demonstrate RNNs deliver superior performance than XGBoost by achieving 94% accuracy, 93% recall, 92% F1-score, and 92% precision, at a time when XGBoost reaches 88.47% accuracy alone. Results demonstrate the advantage of XGBoost-RNN integration for mortality predictions, which enables public health services to create better resource plans.