In this study, we have conducted a comprehensive analysis of blood glucose level prediction using three advanced deep learning models including Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRU). The aim is to evaluate and compare the effectiveness of these models in forecasting blood glucose levels, which is crucial for effective diabetes management. The LSTM model, renowned for its ability to manage long-term dependencies, demonstrates significant potential in managing the temporal dynamics present in glucose data. GRU, known for its efficiency and streamlined version of LSTM, provides a balance between computational demands and predictive accuracy. On the other hand, the traditional RNN serves as a benchmark, offering fundamental insights into basic performance capabilities. To train and evaluate these models, a simulated dataset containing blood glucose measurements was used. Our comparative analysis is centered around critical metrics which is the root mean square error (RMSE). The findings reveal the substantial promise of these deep learning models in augmenting personalized diabetes care and aiding in the advancement of continuous glucose monitoring technologies. From the conducted results, GRU achieved the best RMSE value with 2.74.