<![CDATA[Stunting detection based on anthropometric data requires a classification model capable of capturing complex relationships among child growth characteristics. This study developed a stunting detection model integrating Dual Variational Autoencoder (Dual-VAE), Generative Adversarial Network (GAN), and Transformer using age, sex, weight, and height as predictors. The dataset consisted of 40,071 records of children aged 0–59 months from Jeneponto Regency, Indonesia. Two class-oriented VAE branches were used to learn complementary latent representations of stunted and non-stunted children. GAN-based enrichment was applied to the stunted latent representation, while the Transformer modeled interactions among the resulting latent features. Model performance was evaluated using stratified five-fold cross-validation. The proposed model achieved an accuracy of 0.9960 ± 0.0007, F1-score of 0.9950 ± 0.0008, G-Mean of 0.9957 ± 0.0007, MCC of 0.9917 ± 0.0014, AUROC of 0.9999 ± 0.0001, and AUPRC of 0.9998 ± 0.0001. The results showed that combining class-oriented latent representations, controlled generative enrichment, and feature-interaction modeling produced high and consistent classification performance for anthropometric-based stunting detection.]]>
Copyrights © 2026
