Diabetes Mellitus Prediction using Hybrid Artificial Neural Network

Diabetes mellitus is prominent diseases throughout the world. As human society has become economically and socially isolated, the pervasiveness of diabetes is rapidly expanding. It is the metabolic diseases where a patient has high blood sugar either due to the failure of the body to produce adequate insulin or the failure of the cells to respond to the produced insulin. It can be identified by studying on several readings taken from a patient such as albumin, creatinine, fasting, glucose, potassium, sodium and much more. This study evaluates artificial neural network based algorithms for detection of diabetes. It presents a study on the prediction of diabetes disease through the scaled conjugate gradient back propagation of artificial neural networks using k-fold cross validation. The PIMA Indian Diabetes (PID) dataset has been used from Kaggle. The network is trained using data from 768 diabetes patients aged 21 to 81 years. The accuracy of the results depends on the number of neurons in the hidden layer. The proposed system uses 8 input attributes and provides a minimum percentage of accuracy of 77% and a maximum accuracy of 100% by estimating the presence and absence of diabetes during testing.