Performance Prediction Using Cross Validation (GridSearchCV) for Stunting Prevalence
Taufik Hidayat, Danny Manongga, Hendry Hendry, Yessica Nataliani, Sutarto Wijono, Sri Yulianto Joko Prasetyo, Evi Maria, Untung Raharja, Irwan Sembiring
Abstract
Stunting is a condition in toddlers characterized by inadequate food intake and vitamin deficiency. Predicting stunting in toddlers based on prevalence during data collection from measurement results is crucial for health institutions and related agencies in designing programs. It's an effort to reduce the number of stunted toddlers and allocate resources accurately. It can also help identify toddlers who might need media attention or health program interventions, thus avoiding severe strategic impacts on toddlers' nutritional status that could affect their sustainability and growth. ML, an artificial intelligence component, can forecast stunting in young children by considering ongoing influences on their physical growth and nutritional intake. This research aims to develop several models that provide the best predictions for toddler stunting for health institutions and related agencies. To achieve this goal, researchers developed an approach to predict toddler stunting using four models: Multilayer Perceptron (MLP), RF, Tree, and NB classifiers. To enhance dataset processing, researchers used Grid Search Cross Validation (GridSearchCV) to maximize model performance and select the best parameters to handle imbalanced datasets using SMOTE. This approach yielded a reliable comparison between Cross Validation and GridSearchCV. The Multilayer Perceptron method achieved 99%, Tree 99%, Random Forest 99%, and Naive Bayes 90%.