Evaluating an Ensemble Machine Learning Based Diabetic Medical Illness Analytics Platform for E-Healthcare

Abstract

Diabetes Mellitus (DM) is a glucose metabolism illness that is globally prevalent, long-lasting, slow poison, and a danger to public health. Since there isn't a long-term solution for diabetics, accurate early identification is essential. To determine which machine learning (ML) method predicts slightly earlier mellitus the most accurately, this study compares the Bagging and Boosting Machine Learning models. The algorithms used were RF, Ada, Gradient Boost, XGBoost, some Normal algorithms linear regression, and support vector machines (SVM), all of which exhibited average accuracy. Hence, additional studies are needed to determine diabetic conditions owing to an absence of endeavor and low accuracy. So, to increase the effectiveness of the combined adaptable classification method and reduce the possibility of misdiagnosing a specific example, research work devised an ensemble method known as the "Bagging and Boosting Classifier." Random Forest and XGBoost are employed as conceptual. The proposed Bagging and Boosting classifiers outperform existing models for diabetes diagnosis, such as LR, SVM, Ada, GB, and CatBoost. Accuracy levels for Random Forest (RF) and XGBoost were 99.00% and 99.00%, respectively, which is extremely acceptable. As a result, the accuracy, sensitivity, precision, F1-score, specificity, and ROC AUC metrics are used to evaluate the effectiveness of the Machine learning techniques, and a visual comparison of all the Machine learning techniques is presented as a result. Particularly in impoverished nations, effective diabetes diagnosis using ML systems can considerably lower the yearly death rate. To effectively treat Chronic Diseases, practitioners may benefit significantly from this work. As a result, an effective e-healthcare system can be built in the coming years.