Early Detection of Type 2 Diabetes Using Oversampling-Based Ensemble Learning for Classification of Imbalanced Dataset

Diabetes is marked by high blood glucose levels and adversely affects various body systems, notably cardiovascular, renal, and visual. Early diagnosis via automated screening is vital for operative treatment. Advances in machine learning (ML), particularly have enhanced diabetes diagnosis but one of the major issues when working with medical datasets is class imbalance. Oversampling is crucial in the medical field to address class imbalance, ensuring that models can effectively learn patterns from minority classes, such as rare diseases or conditions. While feature selection and hyperparameter tuning optimize model performance, oversampling directly improves the model's ability to detect and predict critical, but less frequent, medical events, reducing bias toward the majority class. This paper introduces an oversampling based approach for using ensemble learning (EL) with decision tree as meta-estimators to address data imbalance. To balance the classes and enhance the accuracy of classification performance KMeans SMOTE Boost, SMOTE Bagging, Over Boost, and Over Bagging classifier strategies are been used on training data. This study identifies the most effective combinations of oversampling techniques and classifiers, and finds that using oversampling methods consistently leads to better classification results. These results highlight the effectiveness of oversampling with EL system in diabetes classification on Type 2 Diabetes dataset.