Children Food Intake Level Based on Deep Learning Using Variational Autoencoder-Gated Recurrent Neural Network (Vae-Grn2)

Malnutrition in children has been studied in recent years as the insufficiency to consume essential nutrients, resulting in stunted growth, lower immunity, and impaired development. Additionally, according to World Health Organization (WHO) standards, malnutrition is categorized into three major groups: underweight, stunted growth, and wasting. However, training diet controls the accuracy and memory, which is limited by the volume and bias of databases in typical nutritional advice systems. To address this issue, we proposed a Variational Autoencoder-Gated Recurrent Neural Network (VAE-GRN2) framework to predict caloric and nutrient levels in children. Furthermore, the pre-processing of the data obtained from the dataset can be conducted with the help of the C-Score Normalization (CSN) algorithm, which defines errors, duplicates, and missing data. Then, the impact ratios of the various levels of food intake, based on their nutrient profiles, are calculated using the Contrastive Nutrient Impact network (CNIN) model. Additionally, optimal features can be selected using the Deep Variance Sequence Generating Feature Selection (DVSGFS) technique to identify informative nutrients in temporal dependencies and predictive performance. Finally, we proposed a VAE-GRN2 model based on a deep learning (DL) algorithm to improve the accuracy of detecting children’s dietary status by generating feature vectors. Furthermore, the presented methods can detect food intake levels with a high 97% accuracy rate in performance metrics such as precision, recall, F1-score, accuracy, and time complexity.