<![CDATA[Diabetes mellitus is a chronic disease with a continuously increasing number of sufferers. Early detection remains difficult because conventional methods often only recognize the disease at an advanced stage. This study evaluates the performance of the Convolutional Neural Network (CNN) and Multi-Layer Perceptron (MLP) in classifying diabetes using the NHANES dataset (2,278 samples; 21 positive for diabetes). The models were tested with k-fold cross-validation using the metrics accuracy, precision, recall, F1-Score, and ROC-AUC. Results show high accuracy and precision (0.99), an average recall of 0.67, and an F1-Score of 0.75. A paired t-test indicates that CNN is superior in some metrics with a p-value of 0.374, though the ROC-AUC difference is not significant. CNNs can capture complex patterns in health features such as glucose, BMI, and age, whereas MLPs remain reliable as a baseline. In conclusion, both CNN and MLP have the potential to be used for tabular data-based diabetes classification, with CNN showing a tendency to be more effective in detecting non-linear patterns in the imbalanced dataset.]]>
Copyrights © 2026
