Random Forest Optimization Using Recursive Feature Elimination for Stunting Classification

Abstract

Stunting is still a major health problem in Indonesia, with a prevalence of 27% in toddlers in 2023, far from the WHO target of below 20%. RSU Mitra Medika Tanjung Mulia in Medan serves patients with various socio-economic backgrounds, which affects the quality of services, including stunting detection. Conventional methods are prone to bias and error. This study used the Random Forest algorithm and the Recursive Feature Elimination (RFE) feature selection method to improve the accuracy of stunting classification. After data preprocessing and feature selection, two main variables were identified, namely age and height. The initial Random Forest model achieved an accuracy of 94.38%, which increased to 94.42% after hyperparameter tuning. The results showed that this approach produced an accurate, efficient model that can be integrated into clinical systems, helping medical personnel identify children at risk of stunting quickly and accurately, increasing the effectiveness of interventions, and supporting government efforts to reduce the prevalence of stunting