<![CDATA[The demands of workload combined with an unhealthy lifestyle, without being balanced by regular exercise, can lead to diabetes mellitus (diabetes), a disease caused by problems with the pancreas and insulin. Diabetes is the seventh leading cause of death worldwide, while in Indonesia, it ranks as the third-highest cause of death. The objective of this study is to determine the best model to be used for diabetes prediction. Diabetes prediction will be conducted using machine learning with ensemble methods and deep learning. The data is collected from patients at Dr. Kariadi Central General Hospital, where there is an imbalance between positive and negative diabetes cases. The dataset is processed using SMOTE (Synthetic Minority Over-sampling Technique), and the models are configured and optimized using hyperparameter tuning to achieve the best accuracy, precision, recall, and F1-score. The percentage of patients classified as non-diabetic is 91.49%, and this value is used as the baseline for developing the detection system. When building the best model, accuracy alone is not sufficient; instead, precision and recall are more critical for diabetes prediction. Low recall indicates that many diabetic patients go undetected (high False Negative rate), which is dangerous as they do not receive early treatment. Low precision means many healthy individuals are misdiagnosed as diabetic (high False Positive rate), leading to unnecessary anxiety and treatment. The Decision Tree with Ensemble Extra Trees model achieves the highest recall at 99.65%. The Deep Neural Network model produces the highest precision at 99.15%. The best F1-score balance is achieved by the Decision Tree with Ensemble Extra Trees model, with a score of 99.39%.]]>
Copyrights © 2025
