Transforming food authenticity testing by the exploitation of a machine learning - Data fusion approach: a tea case study.

The growing vulnerability of the global food supply chain highlights the need for rapid, accurate, and non-destructive authenticity testing. In this study, we developed and validated a workflow integrating Fourier-transform infrared, near-infrared, and X-ray fluorescence spectroscopy with machine learning-based data fusion for black tea authentication. A total of 532 authentic Assam, Darjeeling, Ceylon, and Keemun samples were analysed using five supervised models. A series of information-level, feature-level, and decision-level fusion strategies were developed and compared, with decision-level fusion achieving 100% F1 scores across calibration, validation, and test sets, outperforming individual and other fused methods. The workflow was further applied to 89 commercial teas, identifying a 6.74% non-compliance rate, all from online platforms. This approach eliminates the need for expensive mass spectrometry or stable isotope-based instrumentation and is well suited for accurate, cost-effective food authenticity testing in non-specialist laboratories, particularly in developing countries.