CNN Performances for Stunting Face Image Classification

Detecting stunting in children is an important issue currently faced by healthcare providers to follow up efforts to reduce the number of stunted children in a region. Emerging research has applied deep learning to many aspects of stunting, including classifying stunting in children through facial images. This study aims to extend previous work on CNN benchmarking in classifying facial images of stunted children and normal children. Here, we trained, validated, and tested five more popular CNN architectures, namely ShuffleNetV2, InceptionV3, MobileNetV2, EfficientNetB0, and VGG19; the research was applied to the same dataset built in previous work. We presented the training and validation loss graphs, and the confusion matrices of all architectures. We measured CNN's accuracy, precision, recall, and F1-measures of the testing stage. Simulation results indicated that three CNN models, namely MobileNetV2, VGG19, and EfficientNetB0 performed comparably or better than previous CNNs in classifying children's facial images. VGG19 and EfficientNetB0 achieved testing accuracies of 100^%.