Malnutrition, particularly underweight, is a critical global health challenge impacting millions of individuals, especially in developing countries. This study presents a predictive model employing machine learning techniques, specifically a neural network, to forecast underweight prevalence using historical nutrition datasets. The research integrates two comprehensive datasets: country-wise averages and malnutrition estimates, to create a robust analytical framework. Through meticulous preprocessing, including data cleaning, feature engineering, and scaling, we derived a new feature, combined_malnutritio, which encapsulates multiple malnutrition indicators. The neural network model, characterized by a single hidden layer with 24 neurons and trained over 100 epochs, demonstrated significant predictive capabilities, achieving a Test Loss of 0.0132 and a Mean Absolute Error (MAE) of 0.0879, indicating an approximate accuracy of 98.67%. The results suggest that the model effectively captures the complexities of malnutrition trends across various countries. Visualizations of the training and validation performance reveal steady improvements, underscoring the model's potential for generalization. This study not only highlights the efficacy of neural networks in public health prediction but also provides a framework for future research into malnutrition dynamics, emphasizing the importance of utilizing multifaceted indicators for better health outcomes.