This Author published in this journals
All Journal Jurnal Teknokes
Ledy Elsera Astrianty
Faculty of Science and Technology, Universitas Teknologi Yogyakarta, Yogyakarta

Published : 1 Documents Claim Missing Document
Claim Missing Document
Check
Articles

Found 1 Documents
Search

GiziLens Application for Toddler Nutritional Status Classification Using SMOTE-NC and Bayesian-Optimized Deep Learning Sherlly Briliant Saputri; Ledy Elsera Astrianty
Jurnal Teknokes Vol. 19 No. 3 (2026): September
Publisher : Jurusan Teknik Elektromedik, Politeknik Kesehatan Kemenkes Surabaya, Indonesia

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.35882/jteknokes.v19i3.183

Abstract

Stunting is a chronic growth disorder in toddlers. Its detection is often hampered by limited conventional diagnostic methods which process the indicators of weight-for-age (W/A), height-for-age (H/A), and weight-for-height (W/H) in a fragmented manner, potentially limiting the understanding of multiple nutritional conditions in toddlers. To address this issue, this research aims to develop GiziLens, a web-based decision support system that integrates the TabNet deep learning architecture using three target-specific models trained separately but executed concurrently on the web application backend to simultaneously produce predictions for the three nutritional status indicators. This study contributes by applying TabNet optimization to local data to handle extreme class imbalance, while transforming it into a practical diagnostic tool. In its implementation, this research utilized 40,071 toddler anthropometric medical records (aged 0–59 months) from Jeneponto Regency for the 2021–2024 period. The class distribution imbalance in the dataset was mitigated using the Synthetic Minority Over-sampling Technique for Nominal and Continuous (SMOTE-NC) to regenerate minority samples until reaching a controlled residual ratio, followed by automatic optimal hyperparameter tuning using Bayesian Optimization through the Optuna framework. The evaluation results demonstrated that TabNet produced a tighter train-test generalization gap compared to MLP (0.32% vs 1.60% for W/A; 1.33% vs 2.06% for H/A; and -0.63% vs -3.56% for W/H). The TabNet model achieved a test accuracy of 96.46% (F1-weighted 0.96) for W/A, 95.88% (F1-weighted 0.96) for H/A, and 98.11% (F1-weighted 0.98) for W/H. The developed GiziLens prototype proved technical feasibility and promising retrospective predictive performance as an initial decision support system prior to conducting prospective clinical trials. The GiziLens prototype holds the potential to assist healthcare workers and program managers at Public Health Centers in accelerating integrated early stunting detection, despite the system was designed as an initial decision support instrument and cannot yet replace independent clinical diagnoses.