Multi-Class Imbalance Classification of Diabetes Cases Using Light Gradient Boosting Machine

Abstract

Diabetes is the third leading cause of death in Indonesia. Diabetes is considered a silent killer because it kills slowly and triggers various complications of chronic diseases in the body of the sufferer. Early detection of diabetes is very important to reduce the risk of more serious health problems and reduce the country's socio-economic losses in diabetes management. Machine learning classification is an alternative method that can be used for early detection of diabetes by predicting category labels from observed data. This study aims to classify diabetes using the Light Gradient Boosting Machine (LGBM) method with Synthetic Minority Oversampling Technique of Nominal and Continuous (SMOTENC). The SMOTENC oversampling method is used to handle the imbalance problem in the dataset used, while the LGBM method is used for multi-class classification of diabetes. The results showed that by applying the SMOTENC technique, a more balanced data distribution was obtained, so that when used in the classification process using LGBM, it resulted in high model performance. Based on the confusion matrix, the accuracy value is 90%.