This work utilizes a multilayer perceptron (MLP) architecture to categorize food items in terms of nutritional density using more than 7000 item data. Pre-processing is conducted on the data by standardizing the features and specifying a binary target variable before the MLP—which includes an input layer, two hidden layers, and a sigmoid output layer—registers 93.69% accuracy and 0.98 AUC. These findings illustrate the efficacy of the model in nutritional profiling, and have implications for personalized nutrition and public health. Problems with dataset diversity and real-world implementation are discussed, and opportunities for future research are highlighted.