Diabetes is a form of metabolic disorder marked by elevated persistent blood glucose (BG), leading to several severe problems in the long term. Continuous monitoring and prediction of BG concentration are needed to help diabetic patients maintain their wellbeing. If insulin is minimized, machine learning models, such as CNN, RNN, and others, are standard data-driven BG prediction solutions. They use several patients' BG data to train the prediction model. However, all of the training data with the same parameters can not accurately capture BG fluctuation characteristics. Motivated by the possibility that various subgroups of diabetic patients have different BG fluctuation trends, we suggest a new BG prediction method called DP-RNN focused on recurrent neural networks (RNN) and incorporates a clustering pre-process using GKFCM. In terms of BG estimation precision, numerical finding shows that the suggested DP-RNN methodology uses more than one cluster for type II diabetes and outperforms Logistic regression (LR) and other CNN approaches.