Food spoilage monitoring has become a critical issue in modern smart supply chains, where safety, transparency, and efficiency are essential. Conventional machine learning techniques often face challenges when handling non-stationary time-series data and distributed environments that demand privacy, adaptability, and scalability. To address these limitations, this study proposes the Federated Attention Latent Temporal Network (FAAS-Chain), a novel federated deep learning framework for intelligent spoilage detection within smart packaging systems. The proposed architecture integrates three major components: (i) Latent Temporal Encoding with Contextual Attention at the edge layer to reconstruct sensor-driven time-series data and detect anomalies, (ii) a federated optimization layer (FedProx) that ensures stable and robust learning across heterogeneous devices while minimizing client drift, and (iii) a secure aggregation layer that maintains data confidentiality through privacy-preserving masking protocols. Together, these components enable FAAS-Chain to provide decentralized, interpretable, and cybersecure monitoring of food quality. Experimental evaluations demonstrate that FAAS-Chain achieves an accuracy of 98.5%, precision of 99.1%, recall of 97.8%, and F1-score of 98.2%, outperforming existing methods such as Random Forests and standalone autoencoders. These results validate the framework’s ability to enhance food safety, minimize waste, and promote sustainable operations. However, future research should address dataset diversity and assess large-scale industrial deployment under varying communication and environmental conditions.
Alrashdi et al. (Thu,) studied this question.