Hospital resource planning is essential for optimizing the allocation and use of resources in healthcare settings. Accurate predictions in this area are difficult due to the complexity of healthcare systems and varying patient demands. Research has explored mathematical models, optimization techniques, and data-driven approaches to improve aspects like capacity planning, workforce scheduling, and demand forecasting. They often fall short in precision and reliability, causing inefficient resource distribution. Hence, there remains a lack of precision in forecasting specific needs such as patient cost categories and bed requirements. This study proposes a framework of enhanced data preprocessing, integrating classifier and regressor models (including Random Forest, Decision Tree, and Neural Networks), and using techniques like SMOTE to handle class imbalance. The models are trained on publicly available data and extensively evaluated with cross-validation techniques. Comparative results with the literature show that the Random Forest Classifier achieved 93.66% best accuracy and the Random Forest Regressor showed a low Mean Absolute Error (MAE) of 0.074 with R2 of 0.9991. These findings suggest that the proposed approach enhances resource allocation accuracy and reliability, aiding healthcare administrators and policymakers.