An Ensemble Learning Framework for Prediction of Diabetes Mellitus using Soft Voting and Hard Voting Framework

Diabetes Mellitus (DM) is a collection of long-term illnesses that arise from excessively elevated blood sugar levels. It ranks among the leading seeds of mortality globally and affects vision loss, kidney issues, heart issues, and other conditions. Machine learning techniques help reduce the workload of medical professionals by improving disease diagnosis and treatment accuracy. Early diabetes prediction could result in better medications. Research in the area of diabetes prediction is mushrooming. Type-1 and Type-2 diabetes are the two fundamental groups into which Diabetes Mellitus divides. The International Diabetes Federation reported 537 million people aged 20–79 with diabetes in 2021, set to rise to 783 million by 2045. To improve the prediction model's efficiency, this research study proposed a diabetes prediction system that makes use of medical data and Ensemble machine learning (ML) algorithms like Naïve Bayes (NB), Random Forest (RF), Adaptive Boosting (AdaBoost), and Gradient Boosting Machine (GBM) with Voting Classification. All of the investigations in this literature were carried out using the Pima Indian Diabetes Dataset (PIDD) i.e. dataset is downloaded from Kaggle. The resultant shows better performance measures in accuracy, precision, recall, F1 score, and AUC with values being 0.8095, 0.7647, 0.6164, 0.6716, and 0.8741 respectively employing Soft Voting classification whereas the values 0.8008, 0.7377, 0.5342, 0.6290 and 0.7291 respectively utilizing Hard Voting Classification.