Prevalence and Early Prediction of Diabetes using an Enhanced Deep Neural Network

At present, diabetes is becoming one of the world's leading causes of adult death from any cause. The dangers of diabetes, especially if it is not managed, include an increased possibility of cardiovascular disease, cancer, renal failure, and vision loss. Timely treatment depends on early diagnosis of diabetes since it can halt the disease's progression if caught early enough. The suggested technique has the potential to aid in not just future diabetes prediction but also in identifying the specific form of diabetes that an individual suffers from. With so many variations between type 1 and type 2 diabetes, this approach will aid in ensuring that the patient receives the most effective care possible. Our model is mostly constructed in the deep neural network's hidden layers and employs dropout regularization to prevent overfitting by re-framing the issue as a classification problem. To produce a highly accurate prediction model from a deep neural network, we tweaked a quantity of attributes and employed the binary cross-entropy loss-function. The suggested DLPD (Deep Learning for Predicting Diabetes) model is shown to be effective and adequate in experimental settings. Train accuracy for the diabetes type dataset is as high as 96.3%, while train accuracy for the diabetes data set from Pima Indians is as high as 99.6%. Research on diabetes and diabetic type in Pima Indians has been extensive. The experimental findings validate the dominance of our suggested design contrasted to the current gold standard approaches.