This study presents a robust methodology to address the challenge of high-dimensionality in clinical datasets, a pervasive issue in biomedical informatics. Leveraging the predictive capabilities of Machine Learning—particularly in healthcare applications—has shown considerable promise. However, the presence of redundant or low-relevance features in training data often hampers model efficiency and inflates both computational and clinical costs. These concerns are especially critical in medical scenarios, where data acquisition may involve expensive or invasive procedures. To mitigate these challenges, we introduce an iterative framework that employs multiple Random Forest models and a Gaussian modeling approach to statistically assess and rank attribute importance. Through repeated training cycles, our approach identifies non-contributory and statistically insignificant features, enabling their systematic exclusion.