Improving Stunting Prediction in Children: Evaluating Ensemble Algorithms with SMOTE and Feature Selection Agus Byna (a*),Fadhiyah Noor Anisa (b), Nurhaeni (a)
a) Information System Department, Science and Technology Faculty, Universitas Sari Mulia, Jl. Pramuka no. 2, Banjarmasin, Indonesia
*agusbyna[at]unism.ac.id
b) Midwifery Department, Health Faculty, Universitas Sari Mulia, Jl. Pramuka no. 2, Banjarmasin, Indonesia
Abstract
Childhood stunting presents a critical challenge to the welfare and health of numerous developing countries, Indonesia included. The phenomenon arises from various factors, including insufficient, excessive, or imbalanced intake of vital energy and nutrients crucial for proper child growth. To address this issue, our study endeavors to develop a predictive model utilizing Machine Learning (ML) techniques. We focus on evaluating three ensemble algorithms on the Banjarmasin Demographic Health dataset to forecast stunting in children under five accurately.
To maximize prediction accuracy, we employ SMOTE (Synthetic Minority Over-sampling Technique) and Feature Selection techniques in conjunction with the three algorithms. By doing so, we aim to enhance the performance of our models and attain the most reliable results. Our dataset comprises 457 instances of stunted children, and we carefully select thirteen pertinent features to incorporate into twelve distinct models.
Upon thorough analysis, we find that the Decision Tree model with SMOTE and Feature Selection emerges as the most accurate, achieving an impressive 90% accuracy score during testing on 70% of the training data. In contrast, the Random Forest model with SMOTE performs less effectively as the weakest predictor for stunting. As a result of our discoveries, we confidently assert that the Decision Tree model with SMOTE and Feature Selection outperforms the other eleven models utilized in this study to predict stunting status among children under five in Banjarmasin.
We intend to expand our research by incorporating more features and data. Additionally, we will explore alternative models, potentially leveraging a combination of Machine Learning and Deep Learning techniques to enhance the predictive capabilities for childhood stunting further. These advancements promise to refine interventions and policy decisions to address this pressing issue and improve the well-being of young children in Indonesia and beyond.
