Addressing Class Imbalance in Medical Data: A Synthetic Minority Oversampling Technique (SMOTE) Approach for Chronic Kidney Disease Prediction

Chronic Kidney Disease (CKD) is a significant health concern in contemporary society. Due to its early asymptomatic nature, it is frequently referred to as the silent sickness. Consequently, the early prediction of this condition is imperative. Nonetheless, a challenge persists regarding class imbalance in CKD datasets, which are huge. Therefore, properly predicting this condition is unfeasible. This research employed SMOTE (Synthetic Minority Oversampling Technique) to balance the dataset and enhance the accuracy of CKD predictions. A range of supervised machine learning methods was employed to predict CKD. We evaluated the performance of machine learning models prior to and after the use of SMOTE. The application of SMOTE improved the performance of the machine learning algorithms. The assessment criteria employed in the models included accuracy, precision, recall, F1 score, and AUC-ROC. Initially, the models exhibited subpar performance in precision, recall, and F1 scoring. Following the use of SMOTE, the precision and recall metrics showed enhancement. The most effective model was XGBoost. It achieved an accuracy of over 96.1%, a precision of 92.6%, and an F1 score of 93.4%. This research will be advantageous for researchers and physicians.