Weight Status Prediction Using a Neuron Network Based on Individual and Behavioral Data

Abstract

Background: The worldwide epidemic of weight gain and obesity is increasing in response to the evolution of lifestyles. Our aim is to provide a new predictive method for current and future weight status estimation based on individual and behavioral characteristics. Methods: The data of 273 normal (NW), overweight (OW) and obese (OB) subjects were assigned either to the training or to the test sample. The multi-layer perceptron classifier (MLP) classified the data into one of the three weight statuses (NW, OW, OB), and the classification model accuracy was determined using the test dataset and the confusion matrix. Results: On the basis of age, height, light-intensity physical activity and the daily number of vegetable portions consumed, the multi-layer perceptron classifier achieved 75.8% accuracy with 90.3% for NW, 34.2% for OW and 66.7% for OB. The NW and OW subjects showed the highest and the lowest number of true positives, respectively. The OW subjects were very often confused with NW. The OB subjects were confused with OW or NW 16.6% of the time. Conclusions: To increase the accuracy of the classification, a greater number of data and/or variables are needed.