Diabetes is a major health concern worldwide, with rising rates and related complications emphasizing the need for effective predictive models. Identifying at-risk individuals early is crucial for managing diabetes and preventing its progression, which is why classification models are essential. This study focuses on improving classification accuracy through an optimized decision tree model that uses a modified grey wolf optimizer (GWO). The GWO was applied to fine-tune the model’s hyperparameters, resulting in the best settings: a max_depth of 35, min_samples_split of 37, and min_samples_leaf of 25. These adjustments led to a significant increase in the model’s performance, achieving an accuracy of 81%, which is higher than other models like Logistic Regression (78.8%), Naive Bayes (76.07%), and support vector machine (78.4%). The model’s effectiveness was further confirmed with a Receiver Operating Characteristic (ROC) curve, yielding an Area Under the Curve (AUC) score of 0.84, indicating its strong capability to distinguish between positive and negative cases. In summary, the findings show that the optimized decision tree, enhanced by GWO, is a promising method for classification tasks and has potential implications for future research in improving model optimization and feature selection techniques.