The class imbalance creates a significant challenge in the development of accurate binary and multiclass clinical prediction models in healthcare. This study compares three different approaches—Random Over-sampling (ROS), Random Under-sampling (RUS), and Synthetic Minority Over-sampling Technique (SMOTE)—to handle imbalanced data in clinical prediction problems. The primary objective of class balancing in healthcare prediction problems is to improve the predictive performance of the minority class, which often represents critical cases. The fine-tuned prediction models are developed using a support vector machine and random forest technique from the balanced training data that was created. Fine-tuning is done through grid search to optimise each classification model. The performance of these methods is evaluated on one binary and one multiclass classification clinical prediction problem. This research highlights the importance of balancing techniques and effective classifier selection to enhance the predictive power of the clinical prediction model.