Classification of Obesity Level Using Deep Neural Networks

This paper presents an estimate of obesity levels in individuals from Mexico, Peru, and Colombia, analyzing their eating habits and physical condition using deep artificial neural networks. The dataset contains 17 attributes and 2111 records, the same are labeled with NObesity (Obesity Level), which allows classification of the data using the following labels: Insufficient Weight, Normal Weight, Overweight Level I, Overweight Level II, Obesity Type I, Obesity Type II, and Obesity Type III. The hyperparameters of the models were optimized using several architectures of deep neural networks. Different feature sets were used to compare the performances of the models. The results, after testing different deep neural network architectures with different numbers of neurons in intermediate layers and learning rates, were: 25% above 90% accuracy, 50% in the range of 80% to 90% accuracy, and 25% below 80% accuracy. The most successful model, in particular, reached an accuracy of 96.46%. This paper contributes to understanding more about obesity prediction and its various levels through the application of deep neural networks based on individual habits, and to know if this model is better than others in machine learning.