PFP-ATNN-QPSO: A Quantum-Optimized Transformer Framework for Harmless Pediatric Food Classification

Abstract

The nutritional safety of children is being jeopardised by the fast expansion of packaged foods that contain complex ingredient formulations. It is difficult to determine the safety of these diets for newborn infants due to concerns with infrastructure and a lack of transparency. Preventing health concerns requires the rapid identification of harmful foods. To address these challenges, we present an Adaptive Transformer Neural Network optimised for paediatric food product suitability via quantum particle swarm optimisation. The EFSA, the FDA, and Health Canada's multi-regional databases are the primary resources for product labels that we use. The preparation of text data is carried out. Hierarchical Semantic Normalisation (HSN) is used in preprocessing to standardise and improve substance names across regulatory vocabularies. Text segmentation follows preprocessing. By utilising age and hazard information, Multi-Scale Regulatory Graph Clustering (MSRGC) categorises substances. Feature extraction is fed via segmented phrases. Using Hybrid Transformer with Adaptive Attention Mechanisms semantic embeddings, we may obtain information about toxicity, allergenicity, and age-appropriateness. To find out if a product is safe or not, the collected features are input into Adaptive Transformer Neural Networks (ATNNs) that have been optimised using Quantum Particle Swarm Optimisation (QPSO). Python carries out the proposed PFP-ATNN-QPSO method. By comparing the PFP-ATNN-QPSO method to existing methods, we find that it outperforms them in terms of accuracy (19.6%), precision (24.8%), and false negative rate (23.6%).