Predicting Macronutrient of Baby Food using Near-infrared Spectroscopy and Deep Learning Approach

Abstract

In Indonesia, malnutrition and overnutrition can still be found especially among children and toddlers who are supposed to get adequate nutrition. The cause of malnutrition is inadequate nutrition intake, both in terms of the quantity and quality of the food. Therefore, mothers need to know whether the food consumed by children has reached the specified nutrition intake recommendation. This research develops a system to predict macronutrient content in baby food using Near-infrared Spectroscopy (NIRS) to obtain the spectral profile of the food and Deep Learning approach such as Deep-belief network (DBN) and Convolutional Neural Network (CNN) to build the prediction model. We used instant-porridge scan data from SCiO to build the model. CNN managed to give the best performance with error for carbohydrate, protein and fat 11.70%, 26.14% and 28.72% respectively.