Diabetes mellitus is a chronic metabolic disorder with a rapidly increasing global prevalence, posing a significant public health challenge. Early detection of diabetes can enable timely intervention and preventive measures, thereby reducing the risk of long-term complications. In this study, a machine learning (ML)-based methodology is proposed for the early prediction of diabetes mellitus. The proposed approach enhances existing prediction systems by improving key performance metrics, including precision, recall, and F1-score, and achieves an efficiency improvement of 4%–10% compared to state-of-the-art methods. Experimental results demonstrate that the support vector machine outperforms other ML algorithms for diabetes prediction, achieving 92% accuracy, 95% precision, 92% recall, 93% F1-score, 92% specificity, and an area under the receiver operating characteristic curve of 0.97.