Development of Toddler Height Prediction Features Based on Convolutional Neural Network for Handling Stunting

Stunting is a significant health problem that affects children's growth, especially in developing countries. Accurate height measurement is essential to identify stunting in toddlers, but conventional methods such as the use of infantometers and stadiometers often face obstacles, especially in terms of process efficiency when dealing with uncooperative toddlers. This study aims to develop a toddler height prediction feature based on Convolutional Neural Network (CNN) algorithm to help handle stunting, with a focus on improving the effectiveness of the toddler height measurement process. The CNN model, which is designed to accurately predict bounding box on the toddler's body in image, allows for the identification of the position and height dimensions of the toddler in the pixels. The results of this prediction are then converted into height measurements by taking into account factors such as the dimensions of the image in the pixels, the distance of the camera, and the angle of view of the camera. The development process involves several iterations of testing, where the accuracy of the measurement results is calculated using Mean Absolute Percentage Error (MAPE). The results of the feature test showed a MAPE of 11.06%, an average accuracy of 88.94%, and an average difference between the actual height and the predicted result of 10.31 cm. With further improvements, it is hoped that this feature can provide convenience in detecting and handling Stunting, especially height measurement.