Diabetes prediction based on risk factors and associated diseases using ensemble machine learning

The aim of this research is to develop a predictive model for diabetes based on risk factors and associated diseases using ensemble machine learning. The problem addressed in this research is to enhance public health and take the correct action. The research emphasizes the need for timely detection and prediction of diabetes to prevent complications and improve public health. The dataset was 253680, and there is an imbalance. After applying the data preprocessing tasks and class balance using random under sampling majority class, there are 70692 instances that were used for the model. The attribute was reduced to 18 from their original 21 features by using feature selection techniques (filter (mutual information and chi-square) and wrapper (recursive feature elimination)). To construct the best proposed model, six experiments were conducted by splitting the dataset into a training, validation, and test set with the ratio of 80%, 10%, and 10%, respectively, using Random Forest, CatBoost, bagging decision tree, AdaBoost, XGBoost, and Extra Tree algorithms. The performance of the model was evaluated using different evaluation parameters such as precision, recall, accuracy, F1 score, AUC, and confusion matrix. The overall accuracies of Random Forest, CatBoost, bagging decision tree, AdaBoost, XGBoost, and Extra Tree are 90.16%, 88.94%, 88.97%, 87.87%, 88.81%, and 89.86%, respectively. Random forest is the best predictive model with an accuracy of 90.16% and ROC of 96% from the others.