Predicting and Associating of Spectral Feature Nutritional Intake Level Using Bayesian Kernel Machine Regression (BKMR) Method

Children's nutrition is an essential determinant of their development, so there is a need to forecast their food consumption adequately. This work presents a model for assessing children's nutritional intake levels from a machine-learning approach. The data cleaning step deals with missing values, outliers, and data inconsistencies to produce high-quality data for analysis from the dataset. Feature selection applies the Level-based Long Short-Term Memory (LLSTM) that captures temporal dependencies in the children's eating patterns. We use Bayesian Kernel Machine Regression (BKMR) for classification since it is not limited to linear relations between food intake and different factors that could influence them. Compared with the other traditional methods, the accuracy of the proposed method is 96.2%. The framework discussed here is located in the field of child nutrition, focusing mainly on the issue of meeting children's basic nutritional needs. This work proves that the probability of nutrition intake can be predicted by the proposed framework, which can help to make better recommendations on children's diet and, therefore, on their future health.