Diagnosis of diabetes mellitus using artificial neural network and classification and regression tree optimized with genetic algorithm

• Published date: 01 July 2020
• Author: Tuncay Özcan,Ebru Pekel Özmen
• DOI: http://doi.org/10.1002/for.2652
• Date: 01 July 2020

RESEARCH ARTICLE

Diagnosis of diabetes mellitus using artificial neural

network and classification and regression tree optimized

with genetic algorithm

Ebru Pekel Özmen

1

| Tuncay Özcan

2

1

Department of Industrial Engineering,

Istanbul University–Cerrahpasa, Istanbul,

Turkey

2

Management Engineering Department,

Istanbul Technical University, Istanbul,

Turkey

Correspondence

Ebru Pekel Özmen, _

Istanbul University -

Cerrahpas¸a, Engineering Faculty,

Industrial Engineering Department,

Annex Building A Block, Floor: 5, 34320

Avcılar _

Istanbul, Turkey.

Email: pkl.ebru@gmail.com

Abstract

Diabetes mellitus is one of the most important public health problems affect-

ing millions of people worldwide. An early and accurate diagnosis of diabetes

mellitus has critical importance for the medical treatments of patients. In this

study, first, artificial neural network (ANN) and classification and regression

tree (CART)-based approaches are proposed for the diagnosis of diabetes.

Hybrid ANN-GA and CART-GA approaches are then developed using a

genetic algorithm (GA) to improve the classification accuracy of these

approaches. Finally, the performances of the developed approaches are evalu-

ated with a Pima Indian diabetes data set. Experimental results show that the

developed hybrid CART-GA approach outperforms the ANN, CART, and

ANN-GA approaches in terms of classification accuracy, and this approach

provides an efficient methodology for diagnosis of diabetes mellitus.

KEYWORDS

artificial neural network, classification, classification and regression tree, diabetes, genetic

algorithm

1|INTRODUCTION

Diabetes mellitus is a major public health problem all

over the world and especially in developing countries.

There were approximately 425 million adults with dia-

betes around the world in 2017, and this number is

expected to reach 628 million by 2045 according to

data from the International Diabetes Federation (IDF).

Early and accurate diagnosis of diabetes is critical for

the prevention of disease and the selection of appropri-

ate treatment. At this point, several statistical methods

and machine learning models have been used to pre-

dict and diagnose diabetes. Classification algorithms

have been widely used for medical diagnosis

(Luukka & Leppälampi, 2006). In the literature, many

studies can be found for the early and accurate diagno-

sis of diabetes mellitus using classification algorithms

(Samant & Agarwal, 2018). These studies can be sum-

marized as follows.

Carpenter and Markuzon (1998) proposed ARTMAP

neural networks for complex prediction problems such

as medical diagnosis. The classification accuracy of the

proposed approach is evaluated on four medical data

sets: Pima Indian diabetes, breast cancer, heart disease,

and gall bladder removal. Chang and Lilly (2004)) used

the variable input spread inference training (VISIT)

algorithm to create a fuzzy system for classification.

The parameters of this algorithm are then optimized

using the genetic algorithm (GA). The performance

analysis of the proposed algorithm is tested on four

benchmark classification problems: iris data, wine data,

Wisconsin breast cancer data and Pima Indian diabetes

data. Gonçalves, Vellasco, Pacheco, and de Souza

(2006) suggested a neuro-fuzzy model using the

Received: 26 May 2019 Revised: 5 November 2019 Accepted: 3 January 2020

DOI: 10.1002/for.2652

Journal of Forecasting. 2020;39:661–670. wileyonlinelibrary.com/journal/for © 2020 John Wiley & Sons, Ltd. 661