Peer-reviewed article
Child Nutrition Prediction for Stunting Prevention Using K-Nearest Neighbor (K-NN) Algorithm
Chandra Kirana, Delpiah Wahyuningsih, Louis Michael ·
Summary
Written by AI, not yet AICS's view · claude-haiku-4-5-20251001This study developed a machine learning tool using the K-Nearest Neighbor algorithm to predict children's nutritional status and detect stunting in Indonesia. The model was trained on nearly 50,000 children aged 0–5 years and classifies nutritional status into five categories using measurements like weight, height, age, and gender. The tool achieved 92% accuracy overall but performed less well for minority categories like malnutrition. The researchers created a mobile application to help health workers assess children's nutrition in real time for early detection and monitoring.
Key points
- K-NN model with k=2 achieved 92% accuracy in nutritional status classification
- Dataset comprised 49,766 children aged 0–5 years from Bangka Belitung Islands
- Model classifies into five categories: good nutrition, poor nutrition, undernutrition, overnutrition, and obesity
- Precision and recall for good nutrition category were 0.94 and 0.99 respectively
- Performance in minority categories like malnutrition was limited due to data imbalance
Limitations
This is a technical validation study of a machine learning model, not an evaluation of the mobile application's effectiveness in practice. The study does not report whether the tool actually prevents stunting or improves child health outcomes when used by health workers in the field.
Abstract
Stunting is a significant public health issue in Indonesia, affecting both children's physical growth and cognitive development. This study aims to develop a child nutritional status prediction application using the K-Nearest Neighbor (K-NN) algorithm as an early detection tool for stunting prevention. The model classifies nutritional status into five categories: good nutrition, poor nutrition, undernutrition, overnutrition, and obesity, using anthropometric data such as age, weight, height, and gender. The dataset comprises 49,766 samples of children aged 0–5 years from the Bangka Belitung Islands Provincial Health Office. The data processing included normalization, feature selection, and k-value testing to optimize model performance. Evaluation results showed that K-NN with k = 2 achieved 92% accuracy, with the best precision and recall in the good nutrition category (0.94 and 0.99). However, performance in minority categories like malnutrition remains low due to data imbalance. The weighted averages for precision, recall, and F1-score were 0.90, 0.92, and 0.90, respectively. This research's novelty lies in integrating the K-NN model into a mobile application, enabling real-time nutritional status assessment for health workers, improving fieldwork efficiency, and facilitating early detection and monitoring.