Product unit neural network trained by an evolutionary algorithm for diabetes disease diagnosis

Abstract

Diabetes disease occurs when the level of glucose in the blood becomes higher than normal because the body is unable to produce the insulin which is needed to regulate glucose. In this study, a new classification method for the diagnosis of diabetes disease was developed. This method is based on a special class of neural network known as product-unit neural networks (PUNN) which was trained by an evolutionary algorithm (EA). We have used EA in order to determine the basic topology of the structure of the PUNN, and to estimate its coefficients weights. The performances of the proposed classifier were evaluated through the sensitivity, the specificity and the classification accuracy using both conventional and 10-fold cross-validation method using the Pima Indian diabetes (PID) dataset. Obtained results reveal that the proposed approach outperforms several famous and recent methods existing in the literature for diabetes disease diagnosis.