AI Model-Based Prediction of Malnutrition Among Children Using Deep Learning Models with Transfer Learning

Child malnutrition remains a pressing global health concern, leading to severe health consequences and mortality among children. Early detection and intervention are crucial to mitigate its impact. Children images contain vital information that can be utilized in the prediction of malnutrition status. In this research work, the effectiveness of VGG16 and SqueezeNet in malnutrition prediction is investigated. Transfer learning is applied to exploit the knowledge of both the pre-trained models in the task of malnutrition prediction. The experiment is carried out in Google Colab notebook. There is scarcity of children images datasets in public domain; therefore, this work uses synthetic images of children to train the models. The dataset is added with images from specific websites which are captured by web scraping script. This research focuses on classifying children into two categories: healthy and malnourished. The performance of both the models is evaluated and the results indicate that both SqueezeNet and VGG16 exhibit promising capability in prediction. The best results are achieved with VGG16 model with Accuracy of 93.73%, F1-value 0.94, Precision of 0.92, and Recall 0.95, and SqueezeNet results are Accuracy of 90%, F1-value 0.89, Precision of 0.99, and Recall 0.81.