Diabetes mellitus is a chronic disease caused by impaired glucose metabolism and has become a global health threat with a steadily increasing prevalence each year. According to WHO and IDF, the number of people living with diabetes is projected to reach 783 million by 2045. This condition demands the development of an accurate and efficient early detection system to support medical decision-making. This study aims to develop an optimized Support Vector Machine (SVM)-based classification model to enhance the accuracy and interpretability of diabetes prediction. The dataset used is the Pima Indians Diabetes Dataset, which consists of eight medical features such as glucose level, blood pressure, and body mass index (BMI). The research stages include data preprocessing, class balancing using the Synthetic Minority Over-sampling Technique (SMOTE), parameter optimization with GridSearchCV, and interpretability analysis through SHapley Additive exPlanations (SHAP). The results show that the optimized SVM model with the Radial Basis Function (RBF) kernel achieved an accuracy of 82%, with a significant improvement in the diabetes class recall value from 0.564 to 0.83 after optimization. The Area Under Curve (AUC) value of 0.871 indicates the model’s effectiveness in distinguishing between positive and negative classes. The SHAP analysis reveals that Glucose, Age, BMI, and Diabetes Pedigree Function are the most influential features in prediction. These findings emphasize that the combination of normalization, balancing, hyperparameter optimization, and interpretability produces a reliable and transparent SVM model. This model has strong potential for implementation in Clinical Decision Support Systems (CDSS) for accurate and explainable early diabetes detection.