<![CDATA[Type II diabetes mellitus (Type II DM) is a public health burden that requires fast and accurate early detection, particularly in primary care settings. Single machine-learning classifiers such as K-Nearest Neighbor (KNN) and Naive Bayes (NB) are widely used but have limitations, including the computational cost of KNN and the strong feature-independence assumption of NB. This study applies an ensemble Voting Classifier (VC) that combines KNN and NB to classify Type II DM using clinical data from 2,390 patients at Mengwi 1 Health Center. Following the CRISP-DM process, we evaluate the models under 80:20 and 70:30 train–test splits using accuracy, precision, recall, F1-score, and ROC/AUC. Compared with the KNN baseline, soft voting consistently improves performance: on the 80:20 split, accuracy increases from 80.33% to 81.59% (+1.26 percentage points) and F1-score from 79.52% to 80.91% (+1.39%); on the 70:30 split, accuracy increases from 80.47% to 82.01% (+1.54%) and F1-score from 79.65% to 81.24% (+1.59%). The soft-voting ensemble also yields higher AUCs, reaching 0.8138 (80:20) and 0.8213 (70:30), and outperforms both single models and hard voting. The novelty of this work lies in demonstrating that a lightweight KNN–NB soft-voting ensemble, designed for the computational constraints of a primary health center and evaluated with repeated cross-validation, can provide small but consistent gains over single classifiers on real DM data. These findings indicate that such an ensemble is a promising building block for clinical decision support in resource-limited primary care, although further calibration, external validation, and prospective testing are still required.]]>
Copyrights © 2025
