Diabetes is one of the most significant non-communicable diseases and can cause serious complications in individuals such as heart disease, kidney failure and blindness. Disease identification and staging at an early and specific level is of outmost importance in order to avoid further deterioration of the patient’s health. This research aims at developing a more complex classification method for diabetes prediction through a two-stage model. The initial process involves using the Support Vector Machine to select the features from the dataset to help boost the accuracy of the model. In the second phase, a Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) is used to classify the dataset, thus enabling capturing temporal pattern in data related to diabetes. One of the major strengths of LSTM-RNN for electricity load forecasting is its ability to handle missing data and irregular time series data as it produces better predictive results as compared to other models. Our method shows superior values in the measures of correctness, specificity, capacity, and stability in terms of data perturbation taking into consideration of non-linearity inherent in the diabetes dataset. The study points to the possibility of the combination of SVM and LSTM-RNN to enhance the diagnosis and treatment of diabetes.