Diabetes Multiclass Prediction Using Ensemble Learning Techniques

Diabetes is one of the most prevalent diseases in the modern era, leading to a significant number of deaths annually, as reported by the World Health Organization. Early prediction of diabetes can substantially improve patient outcomes and save lives. This study introduces a new model for predicting diabetes using the Random Forest algorithm, known for its powerful ability to split data until reaching an optimal state. Two datasets are utilized: the Multiclass Diabetes dataset and the PIMA Indian Diabetes dataset. The data are preprocessed by removing outliers, handling missing values, and balancing the classes. These preprocessed data are then classified using the Random Forest algorithm through continuous splitting until the stopping criteria are met, aiming to predict diabetic individuals. The proposed model demonstrated superior performance with the Multiclass Diabetes dataset, it achieves a validation accuracy of 100%, a precision of 98.20%, and recall and F1 scores of 98.11% and 98.12%, respectively. With the PIMA dataset, the proposed model achieves a validation accuracy of 85.30%, with precision, recall, and F1 scores of 88.07%, 87.50%, and 87.50%, respectively. In addition to our proposed model, we built many machine learning models with the first dataset such as SVM, logistic regression, logistic regression with L1/ L2 regularization, K-NN, and naïve bayes. Our results indicate that the Random Forest algorithm significantly outperforms other machine learning techniques in predicting diabetes, offering a highly accurate and reliable tool for early diagnosis. This research underscores the potential of ensemble learning in healthcare, particularly in managing chronic diseases like diabetes.