<![CDATA[Once viewed predominantly as a disease of adulthood, diabetes mellitus has become an escalating concern among Indonesian children and adolescents, with type-1 diabetes cases in the under-18 cohort rising approximately seventy-fold between 2010 and 2023. Against this backdrop, this study constructs a web-based clinical intelligent system that harnesses the Support Vector Machine (SVM) algorithm for early risk identification in patients aged 6–18 years. Unlike prior SVM-based diabetes detection studies, which have largely relied on adult benchmark datasets and treated the problem as standard binary classification, this study assembles a pediatric-specific dataset of 500 medical records, comprising 350 clinical records (70%) and 150 re-screened public records (30%), with 10 clinical features. The observed class imbalance (43% positive, 57% negative) is addressed using Synthetic Minority Over-sampling Technique (SMOTE), applied solely within the training partition, while feature thresholds are adjusted to WHO pediatric standards. Data preprocessing includes handling missing values, Min-Max normalization, and label encoding. The SVM model with a Radial Basis Function (RBF) kernel was optimized using Grid Search with 5-fold cross-validation, yielding optimal parameters of C=10 and gamma=0.1. On a held-out test set of 97 records, the model achieved 84.54% accuracy, 81.82% precision, 83.72% recall, and an 82.76% F1 score. The accompanying web application, developed using Python Flask and Bootstrap 5, passed all functional black-box tests. Targeted at frontline healthcare workers in primary care settings rather than lay users, the system provides a practical point-of-care screening instrument for clinicians managing pediatric diabetes risk.]]>
Copyrights © 2026
