Data-Driven Prediction of Microbiological Quality in Oregano-Active Packaged Gilthead Sea Bream Fillets using Non-Invasive Spectroscopy and Machine Learning

S-producing bacteria were identified as the dominant spoilage organisms, with higher temperatures accelerating growth. Vacuum packaging effectively delayed spoilage compared with aerobic storage. The EO treatment resulted in limited antimicrobial effects but influenced sensory characteristics and spoilage-associated spectral responses. Machine-learning regression models were developed to predict total viable counts (TVC) from multispectral imaging (MSI) and FTIR data. Model performance varied across datasets, with both MSI and FTIR enabling the development of predictive models. Linear models provided stable and consistent performance, particularly after appropriate preprocessing, while non-linear approaches captured more complex relationships in certain cases. Importantly, optimized preprocessing and modelling pipelines substantially improved predictive performance compared with raw spectral data. Overall, MSI and FTIR, combined with tailored machine-learning approaches, represent practical, non- destructive tools for real-time monitoring and prediction of microbial spoilage in gilthead sea bream fillets, highlighting both the capabilities and limitations of ML-assisted spoilage prediction in complex seafood systems.