Explainable Adaboost Model for the Early Prediction Of COVID-19 Mortality

The viral COVID-19 pandemic has strained the healthcare sector due to the massive influx of infected patients, putting hospitals and medical care units in a critical state where hospitalization demand exceeds capacity. Over 700,000 positive cases have been reported in KSA, with deaths exceeding 8,900. Although COVID-19 has a relatively high survival rate, this may decrease when there is a risk of deterioration in patients during hospitalization. Therefore, it is essential to invest in health systems to allocate medical resources efficiently and improve patient care. This study aims to employ ensemble-based machine learning models Random Forest (RF), AdaBoost (AB), CatBoost (CAT), and Light Gradient Boosting (LGB) to predict COVID-19 patients’ mortality early. Furthermore, the Explainable Artificial Intelligence (EAI) technique was used to enhance trust by integrating interpretability and comprehensibility into the black-box Machine Learning (ML) models. Hospitalized patients at King Fahad Hospital of the University (KFHU), Khobar, Kingdom of Saudi Arabia (KSA), between March 01, 2020, and February 28, 2021, were included, and a comprehensive dataset of 399 COVID-19 cases was obtained. Three experiments were performed, using all features, employing Sequential Feature Selection (SFS), and a chi-square (Χ²⁾ filter to identify the most significant features for predicting mortality. Experimental results indicate that AB outperformed other classifiers with accuracy, sensitivity, specificity, and F-measure of 0.969, 0.976, 0.938, and 0.982, respectively, with the selected k-best features. It is concluded that the model can aid medical staff in predicting the disease severity of COVID-19 patients to support efficient patient management.