Nowadays, the diabetes mellitus disease is increasing rapidly across the world. It is a chronic metabolic disorder, where patients unable to manage their blood glucose levels will lead to heart attack, blindness, renal disease, kidney problems etc. and that causes a lot of deaths per year. Patients need to pay a lot of money to practitioners and medicine for the maintenance of blood glucose level. So, it is necessary to develop a smart and intelligent system based on artificial neural network that can help network early detection of diabetes and be helpful to the practitioner for the early treatment so it can stop the progress of the disease. In this work, a dataset from Pima, India was taken to predict the possibility of diabetes. The objective of the paper was to minimize the error function in neural network training using a neural network model. The average error rate decreases during training and gave an accuracy of 95.93% for the predication of whether a person is diabetic or not.