Object Segmentation in Stunted Face Images using Deeplabv3+ with Resnet-50

Stunting is the impaired growth and development that children experience from poor nutrition, repeated infection, and inadequate psychosocial stimulation. This study explores the impact of data preprocessing, specifically using DeepLabV3+ segmentation, on the performance of ResNet-50 in classifying stunting and non-stunting facial images. Initially, ResNet-50 achieved 99% accuracy and a 3.22% loss with the unsegmented dataset. By applying DeepLabV3+ to remove irrelevant features and backgrounds, the model's performance improved to a perfect 100% accuracy and a reduced loss of 0.45%. These results underscore the importance of high-quality data preprocessing in enhancing model precision and reliability. The findings have significant implications for practical applications, particularly in medical imaging, where improved diagnostic accuracy can benefit patient outcomes. Further research is recommended to explore additional preprocessing methods and their effects on model performance across diverse domains. This study highlights the transformative potential of effective data preprocessing in optimizing deep learning models for more accurate and reliable machine learning solutions.