Diabetes prediction is a critical application of machine learning, enabling early diagnosis and improved risk assessment. This study evaluates and compares four classification models-Decision Tree, Support Vector Machine, Random Forest, and Multinomial Logistic Regression-using real-world patient data. The dataset was split into training and testing sets, and model performance was assessed using accuracy metrics. Random Forest emerged as the most accurate model, demonstrating superior predictive capability. Feature importance analysis identified HbA1c as the most influential predictor, confirming its strong correlation with long-term blood glucose levels. BMI ranked second, highlighting its association with obesity and increased diabetes risk. Age was the third most significant factor, reflecting the higher prevalence of diabetes in older populations due to declining insulin sensitivity over time. These findings emphasize the value of machine learning in identifying key risk factors and improving diabetes prediction, supporting data-driven decision-making in healthcare analytics.