Performance Evaluation of a Remote Diabetes Healthcare Disease Prediction Framework Using Machine Learning Paradigm for e-health Services

Abstract

Diabetes is a group of metabolic diseases caused by chronically increased glucose levels and is frequently referred to as a chronic illness. Also, diabetes is caused by an imbalance in different parameters like BMI, insulin, blood pressure, skin thickness, and age. It can lead to an infection, heart problems, eye problems, gastric problems, dental problems, foot and hand problems, kidney problems, high blood pressure, and damage to other organs in the human body. If accurate early prediction is possible, the risk factor and severity of diabetes may be reduced significantly. In this paper, we propose a robust framework for diabetes prediction that employs Logistic Regression (LR), Support Vector Machine (SVM), K-Nearest Neighbor (KNN), Artificial Neural Network (ANN), Decision Tree (DT), Random Forest (RF), MLP Classifier (MLP), Gradient Boosting Classifier (GB), Ada Boost classifier (ADA), Nave Bayes (NB), Cat boost classifier (CAT), XG Boost classifier (XGB), Gaussian Process Classifier (GP), Quadratic Discriminant Analysis (QD), and LightGBM (LGBM) classifiers. In this study, it is also indicated that the weighted assembly of several ML models, where the weights are determined using the corresponding Area Under ROC Curve (AUC) of the ML model, improves diabetes prediction. All of the research in this work was conducted using the Pima Indian Diabetes Dataset under similar experimental conditions. With sensitivity, specificity, false omission rate, diagnostic odds ratio, and an AUC of 0.776, 0.925, 0.091, 66.281, and 0.937, respectively, our proposed ANN classifier is the best performing classifier from all the extensive experiments. The accuracy achieved by the Artificial Neural Network (ANN) approach was found to be 94%. Our findings show that artificial neural networks (ANN) outperformed other machine learning techniques in terms of accuracy.