Early Prediction of Diabetes Disease in Women with Tree-Based Classification Algorithm

Abstract

One of the chronic diseases that continues to increase in the world is Diabetes, with Indonesia ranked fifth in the number of diabetes sufferers, especially in women. Early diagnosis is important to reduce the risk of serious complications, but is often hampered by late identification of symptoms. In supporting early diagnosis of diabetes in women, a tree-based classification is proposed, namely Decision Tree, Random Forest and XGBoost. Normalization with min-max is applied to the data and Stratified cross validation with k = 10 fold is applied when creating the model. The results showed that XGBoost had the best performance with accuracy, precision, recall, and F1-Score of 0.992, followed by Random Forest and Decision Tree with an evaluation value of 0.991. This study indicates that XGBoost has proven to have advantages in prediction accuracy, as shown in the confusion matrix with prediction accuracy levels of 99.5% and 98.7%, respectively. The classification error that occurs is very small, making it a reliable solution to support the early diagnosis system for diabetes. Based on these findings, XGBoost is proven to be more reliable in early prediction of diabetes in women compared to other comparative algorithms, namely Decision Tree and Random Forest. This study provides a significant contribution to data-based predictive technology in Health with the potential for wider application.