Detection of Malnutrition in Children Using Deep Learning Model

Early detection of malnutrition in children allows for timely intervention and treatment, preventing further deterioration of their health and growth, and minimizes the long-term consequences of malnutrition, such as stunted growth, impaired cognitive development, and increased vulnerability to diseases. This paper uses ResNet 18 model for identification of malnutrition in children using transfer learning. The contribution of this work includes preparation of children dataset comprising malnutrition and nutritious children. This work presents performance of ResNet18 model for classification of children images. Experimentation is carried out by using various optimizers and different values of learning rate during model training. The model is trained on image dataset constructed by cleaning images generated by stable diffusion model. The model is tested on randomly selected child images from websites. ResNet18 model gave satisfactory results when trained with learning rate 0.0001 for Adam, AdamW, Adamz, RMSprop, and SGD. However, model gave optimal performance for Adamax with precision 100%, Recall 100%, F1 score 100%, train accuracy 91.76%, and validation accuracy 100%.