Implementasi Algoritma XGBoost untuk Klasifikasi Stunting pada Anak di Kota Medan

Abstract

Machine learning (ML) does not rely on explicitly written programming instructions. In the healthcare sector, ML is widely used for classification tasks, including the classification of stunting risk in children. This study develops a stunting risk classification model using the XGBoost algorithm, which is effective for large and complex datasets. The model predicts children’s nutritional status based on anthropometric data such as age, weight, and height, as well as social factors. Model evaluation shows an accuracy of 97.23% on the training data (12,143 samples), 95.69% on the validation data (3,036 samples), and 76.73% on the testing data (1,066 samples). The evaluation metrics indicate good overall performance, although classes with smaller sample sizes present challenges. The Mild class achieves a precision of 0.57, recall of 0.51, and F1-score of 0.54; Moderate has a precision of 0.58, recall of 0.78, and F1-score of 0.67; Normal reaches a precision of 0.83, recall of 0.88, and F1-score of 0.86; Severe shows a precision of 0.90, recall of 0.74, and F1-score of 0.81; and Tall records a precision of 0.97, recall of 0.59, and F1-score of 0.74. Analysis of the loss and accuracy curves indicates that training accuracy reaches 97.23%, while validation accuracy plateaus at 95.69%. The small gap between the curves suggests good generalization capability. Feature importance results show that physical variables such as weight (0.22), height (0.19), and age (0.13) contribute the most, followed by gender (0.14). Social and environmental factors, including housing conditions (0.05), socioeconomic score (0.03), parental occupation (0.02), and breastfeeding (0.02), also play crucial roles in determining children’s nutritional status. This study demonstrates that the XGBoost model is effective in identifying stunting risk and supporting data-driven public health policies.