Pediatric Diabetes Prediction Using Deep Learning

Abstract

The present study proposes a novel technique for the early prediction of diabetes with the utmost accuracy. Recently, the contemporary methodologies of artificial intelligence and in particular Deep Learning (DL), have proven to be expeditious in the diagnosis of diabetes. The model that is supported has been constructed with the implementation of two hidden layers and a multitude of epochs of Deep Learning Neural Network (DLNN) utilizing the Multi-Layer Perceptron (MLP) technique. We proceeded to meticulously adjust the hyperparameters within the fully automated DLNN architecture, with the aim of optimizing data pre-processing, classification and prediction. This was accomplished by a novel dataset of Mansoura University Children's Hospital Diabetes (MUCHD), which allowed for a more comprehensive evaluation of the system’s performance. The system is validated and tested on a sample of 548 patients, each exhibiting 18 significant features. Various validation metrics were employed to ensure the accuracy and reliability of the results like K-folds, leave-one-subject-out and cross-validation approaches with various statistical measures of accuracy, f-score, precision, sensitivity, specificity and dice similarity coefficient. The high-performance level of the proposed system can help clinicians to accurately diagnose health and different diabetes grades with a remarkable accuracy rate of 99.8%. According to our analysis, the implementation of this method results in a noteworthy increase of 4.15% in overall system performance when compared to the current state-of-the-art. As such, we highly recommend the utilization of this method as a promising tool for forecasting diabetes.