ARTIFICIAL INTELLIGENCE FOR PREDICTING CHILD STUNTING RISK

• Main Author: Nur Kamila, Aisya
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/66832
• Institution: Institut Teknologi Bandung

ABSTRACT ARTIFICIAL INTELLIGENCE FOR PREDICTING CHILD STUNTING RISK By Aisya Nur Kamila NIM: 18318037 (Undergraduate Program in Biomedical Engineering) Stunting is a serious nutritional problem in Indonesia. Based on the Indonesian Toddler Nutrition Status Survey, in 2019 the stunting prevalence rate in Indonesia reached 27.67% with East Nusa Tenggara Province having the highest prevalence of 43.8% and Bali Province having the lowest prevalence of 14.4%. According to WHO (2018), this figure belongs to the classification of high stunting prevalence, which is in the range of 20-30%. In the long term, stunting can affect national productivity. Since the effects of stunting are generally permanent after it occurs, it is not necessary only to know about stunting, but also to predict the possibility of stunting before it happens, so that optimal preventive measures can be implemented. Currently, the majority of identification is only detection, not prediction. Whereas with early predictions, stunting prevention programs are optimized. For this reason, in this study, a prediction system will be designed by utilizing machine learning. Data is used from the Klinik Rumah Bersalin Cuma Cuma (RBC) in Bandung, a family with a lower-middle socioeconomic background. Two datasets, child polyclinic data and socioeconomic data, were taken into one. The data is prepared and explored first. After that, the dataset was divided randomly into the training dataset and the test dataset. The training dataset is trained with classification models using the available Scikit-Learn library. The best accuracy is 89.77% with the default parameter Random Forest algorithm trained with the SMOTE oversampling dataset. The most significant features in predicting stunting risk were month distance, body weight, and body length. Socioeconomic features are not very significant but still have an influence. The model with the best accuracy is extracted to a web application designed with Flask and the Heroku server so that health workers can use the stunting risk prediction system that has been created. Keywords: stunting, prediksi, machine learning, child data, socioeconomic data.