Use of Artificial Neural Network Models for Prediction of Diabetes Mellitus

Abstract

Background: Diabetes Mellitus is a chronic metabolic disease that is defined by elevated levels of blood glucose because of insulin insufficiency or insulin resistance. Around 537 million adults worldwide have diabetes, and India appears to be a diabetes hotspot, so it's critical to detect the disease early in order to manage and avoid complications.

Objective: To develop and compare the predictive performance of three Artificial Neural Network models—Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Feedforward Neural Network (FNN).

Methods: A study was conducted among 800 patients, including 400 diabetic and 400 non-diabetic, visiting the hospital outpatient department. Data were collected from demographic, lifestyle, anthropometric and medical history. Qualitative data have been coded and normalization has been done for the quantitative data. The dataset was divided into 80% for training and 20 % for model testing. Accuracy, sensitivity, specificity and Kappa statistics were used to compare the performance of the models.

Results: Risk factors that were found to be statistically significant included body mass index, waist circumference, neck circumference, stress score, diet, and family history of diabetes, cardiovascular disease, or parental high blood pressure. MLP demonstrated the highest accuracy (81.88%) compared to CNN (73.33%) and FNN (75.62%). The MLP was also superior in sensitivity (81.25) and kappa score (0.63) compared to the other models.

Conclusion: MLP has been shown to be the best artificial neural network model for predicting diabetes, which may demonstrate the importance of an artificial neural network for early risk detection.