Identifying malnutrition from facial features of children using deep dearning

Abstract

Malnutrition remains a significant global health issue, particularly affecting children, leading to severe developmental challenges. Current diagnostic methods are often inefficient, relying on time-consuming physical assessments. This study explores the potential of deep learning techniques to automate the detection of malnutrition in children by analyzing facial images. Several state-of-the-art Convolutional Neural Networks (CNNs), including VGG16, VGG19, InceptionV3, MobileNetV2, DenseNet121, and a custom MobileNetV2, were trained and evaluated on a dataset of 1,953 images. After augmentation, the dataset increased to 9,765 images to enhance the models’ performance. Among these models, the custom MobileNetV2 achieved the highest accuracy, with 99.23% during training and 98.57% on validation data, significantly outperforming other architectures. This demonstrates the effectiveness of using CNNs for malnutrition detection. The results highlight the potential of AI-driven solutions to offer faster and more accurate assessments, which could be particularly beneficial in resource-constrained settings. Future research will focus on improving model robustness by incorporating a more diverse dataset, additional health indicators such as body mass index (BMI), and further optimization for mobile application deployment. These advancements could help healthcare professionals to detect malnutrition early and provide timely interventions to improve child health outcomes.