MACHINE LEARNING IMPLEMENTATION IN PREDICTING STUNTING ON CHILDREN AGED 5 AND UNDER USING INDIA’S DEMOGRAPHIC AND HEALTH SURVEYS 2019-2021

• Main Author: Zharif Fadhil, Muhammad
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/75858
• Institution: Institut Teknologi Bandung

Anthropometry is one of the important indicators for the health of infants and children during the early growth period, specifically under the age of 5 years. Stunting is a term used to describe the condition of the human body when growth is inhibited due to malnutrition, resulting in a child or infant having a height below the standard. Stunting, determined by a child's height with a standard deviation of -2 SD from the height-for-age, can lead to difficulties in cognitive development, susceptibility to diseases, lower social tendencies, and various morbidities up to mortality. Based on the Indonesian Nutritional Status Survey, the prevalence of stunting in Indonesia in 2021 was still at 24.4%, indicating a high rate of stunting in the country. This shows that Indonesia needs to intervene and prevent stunting. Stunting is fundamentally caused by various factors related to the child and is not limited to the child's nutrition and anthropometry. In this regard, prevention efforts against stunting can utilize artificial intelligence, particularly machine learning, to process data and identify various stunting factors while creating a predictive model of risk classes based on these factors. In this research, data exploration was conducted using the Demographic and Health Surveys from India due to the similarities and availability of data in Indonesia. Data processing was done to adjust to the Indonesian context, and it was found that the most influential features for training the predictive model of stunting risk classes in children under 5 years old were maternal health, health, intellectual, and anthropometric factors, family socioeconomics, living environment, and child nutrition intake. Through the process of training the predictive model, a K-Nearest Neighbour-based model was found to be optimal in predicting the condition of stunting in children. Further optimization was done by adjusting hyperparameters of the model, resulting in an accuracy rate of 91.183%. The most optimal model was then implemented through a simple website-based application using the Flask microframework.