Diabetes is a widespread health issue affecting millions globally, with an especially high prevalence in Bangladesh. It occurs when an individual experiences persistently elevated blood glucose levels and can lead to serious complications such as vision loss, kidney failure, heart attacks, and strokes. Detecting the condition at an early stage can significantly improve patient outcomes and potentially save lives. However, the incidence of diabetes continues to grow at an alarming rate. This study aims to assess the predictive capabilities of several commonly used machine learning algorithms in identifying early-stage diabetes. With recent progress in the field of machine learning, healthcare applications have seen notable improvements in diagnostic accuracy. In this research, six classification algorithms were applied: Gaussian Naive Bayes, Random Forest, XGBoost, Logistic Regression, Decision Tree, and a Voting Classifier. The analysis was conducted on a real-world dataset consisting of 4,807 patient records-3,875 diabetic and 972 non-diabetic instances. Among all models tested, the Random Forest algorithm achieved the highest accuracy at 98.56%, indicating its strong potential for effective diabetes risk prediction compared to other methods.