<![CDATA[Diabetes was a chronic disease that continued to increase globally, making early detection essential to reduce long-term complications. This study aimed to develop a desktop-based diabetes prediction system that provided fast and simple classification results for medical personnel and individual users. The system used the entropy-based subset selection (E-ss) method to choose the most relevant attributes and a classification tree to classify the risk. The dataset from the National Institute of Diabetes and Digestive and Kidney Diseases, contained 768 patient records with attributes such as number of pregnancies, glucose level, blood pressure, and other risk factors. The E-ss process produced three attributes with the highest information scores, namely body mass index (BMI), blood pressure, and triceps skinfold thickness. These three attributes were then used as input to the classification tree model to generate diabetes risk predictions. Cross-validation testing showed an accuracy of up to 78.95%. These findings indicated that E-ss feature reduction helped maintain prediction performance while improving computational efficiency. This system was expected to serve as a practical and reliable diagnostic tool. ]]>
Copyrights © 2025
