In the wake of the COVID-19 pandemic, efficiently allocating ICU resources for critical patients has become crucial, especially for those with chronic conditions. This study harnesses machine learning (ML) to forecast ICU admissions among COVID-19 patients in Kuwait, analyzing a dataset of 4399 patients to identify pivotal predictors for ICU needs. Employing cross-validation and Synthetic Minority Over-sampling Technique (SMOTE) to tackle data imbalance, the predictive variables were refined using backward feature selection with logistic regression and evaluated model interpretability with Shapley additive explanations (SHAP). The Support Vector Machine (SVM) model outperformed other models with an area under the curve (AUC) of 0.91, and the Extra Tree (ET) model showed better performance with an accuracy of 96.42%. Critical predictors included demographics, clinical outcomes like shortness of breath, elevated d-dimer levels, and abnormal chest X-rays. This research not only underscores the potential of ML in critical healthcare decision-making during pandemics but also highlights its role in discovery science, suggesting broader applications in healthcare and other scientific domains. The study advances medical informatics by integrating ML with healthcare, offering insights into disease dynamics and improving resource allocation strategies.