Application of Artificial Neural Networks for Early Detection of Diabetes Mellitus: Insights from a Case Study in Kaura Namoda Nigeria

Abstract

Diabetes Mellitus (DM) is a chronic disorder which needs urgent attention. Detection of the disease from the grass root using patient risk factors is the key for early prevention of the disease. The aim of this paper is to use Artificial Neural Network (ANN) for early detection of DM.  Datasets were collected from past patient records of patients suffering from DM in General Hospital Kaura Namoda, Nigeria between 2019 to 2023. The hospital and region are selected because most of the cases of DM were reported in the hospital and the prevalence of the disease in the region is about 8% to 10%.  The datasets consists of two sample size of 400 patients each, the first sample dataset consist of 400 patients with demographic, clinical and lifestyle risk factors and second sample dataset consist of 400 patients with demographic, clinical, lifestyle and dietary risk factors. Backward stepwise feature selection method was employed to eliminate the least informative features and the method retained Six (6) risk factors for the first sample dataset age, family history of DM, blood glucose level, blood pressure level, body mass index, physical activity, and removes one risk factor sex. For the second sample dataset, the method  retained twelve (12) risk factors age, family history of DM, blood Glucose level, blood pressure, body mass index, physical activity, preference for sweet food, red meats, refined carbs, energy drinks, white rice, processed meat and remove sex  and preference for salty food. The results of the analysis showed that MLPNN model demonstrated high accuracy in detecting DM and non-DM patients, with improved performance when dietary risk factors were included. The paper concludes that in order to detect DM and Non-DM accurately, dietary risk factors must be included apart from demographic, clinical and lifestyle risk factors.