Hybrid Deep Learning and Anthropometric Model for Efficient Malnutrition Detection in Resource-Limited Settings

One serious problem for public health is malnutrition, particularly among children, as it affects growth, development, and overall well-being. In this project, we propose a hybrid architecture combining MobileNetbased image classification and general mathematical calculations to improve the detection of malnutrition in children. The MobileNet model, pre-trained on a large dataset and fine-tuned on a dataset of malnutrition-related images, is utilized to classify children as malnourished or healthy based on visual cues. Additionally, general mathematical methods, incorporating anthropometric measures such as height, weight, and the Mid-Upper Arm Circumference (MUAC), are integrated to enhance classification accuracy. The hybrid approach outperforms traditional ResNet-based systems, achieving a loss of 0.1883, an accuracy of 94.76%, and an F1-score of 0.9092. The system's effectiveness is further validated by analyzing the confusion matrix and the training loss graph. This novel fusion of mathematical and machine learning methods offer a reliable way to identify malnutrition, enabling its deployment in environments with limited resources. via an interface based on Streamlit. The proposed system aims to support healthcare professionals in early diagnosis and intervention, contributing to the fight against childhood malnutrition.