An Innovative Approach of Early Diabetes Prediction using Combined Approach of DC based Bidirectional GRU and CNN

These days, diabetes is a worldwide epidemic that has reached epidemic proportions in Bangladesh. Blindness, renal disease, kidney issue, heart disorders, etc., are only few of the many deaths that can be attributed to this blood poisoning. That’s why it’s crucial to figure out a way to accurately diagnose diabetes from a patient’s medical history. One of the most dangerous and persistent causes of persistently elevated blood sugar levels is diabetes. Untreated and undiagnosed diabetes leads to a host of consequences. Due to the time-consuming nature of identification, patients often end up in diagnostic clinics for consultations with medical professionals. However, the development of DL techniques has addressed this serious issue. The purpose of this dissertation is to provide early and accurate diabetes diagnoses to medical practitioners through the application of DL techniques. The suggested method utilizes preprocessing, PCA-based feature selection, PCA and itself to train the model. The proposed method uses DC-BiGRU-CNN to train the model. This uses a multigranularity CNN and an attention strategy to accomplish its dual-channel CNN + BiGRU goal. The proposed method was compared to the commonly used CNN and the GRU models. The proposed method has superior performance to both of the competing models.