Accurate blood glucose predictions are crucial for the management of diabetic health. As such, effective management is required to prevent complications. This study proposes a personalized machine learning framework to improve the accuracy of blood glucose forecasting for both short-term and long-term intervals, which utilizes continuous glucose monitoring data. The research employs a sliding window approach and optimization of the hyperparameters using the Optuna framework. Four machine learning models were deployed in the study—convolutional neural networks, long short-term memory, artificial neural networks, and XGBoost. Recursive forecasting was implemented to predict blood glucose levels at multiple time points, ranging from 15 min to 2 h, with a particular focus on improving long-term forecasts. Previous research focused primarily on short-term predictions, leaving a gap in longer-term forecasting. Compared to existing work, XGBoost outperformed all models, achieving an RMSE of 7.47 mg/dL for the 30 min horizon and 17.11 mg/dL for the 120 min horizon—significant improvements over previous studies. This demonstrates the ability of the proposed framework to fill the long-term forecast gap, improving health management for diabetic care.