Diabetes Prediction Using Ensembled Machine Learning Models

Diabetes is a state in which the sugar level in the blood rises due to inadequate supply of insulin in the pancreas or inability of the body to use the insulin effectively. The situation in which the blood sugar level increase to an acute level is called Hyperglycemia which leads to damage of blood vessels. Around 22 % of global deaths during 2000 to 2019 were caused by diabetes. Approximately 1.5 million deaths occurred in the 2019 due to diabetes. Such statistical analysis have led to the evolution of many machine learning based predictive models for the early diagnosis of diabetes in people. This project presents a comparative analysis of different ML models including K-Nearest Neighbors, XGBoost, Logistic Regression, Support Vector Machine, Adaboost, Decision Tree, and Random Forest. Further, hybrid models are generated using bagging and boosting ensembling techniques with each of these weak learners as base models. All these models were compared with respect to the accuracy, precision, recall, specificity, and ROC metrics. It was noticed that the SVM model with the boosting ensembling technique gives the best accuracy of 79%, F1-score of 79% and an AUC value of 85.