Multi-Modal Fusion With Machine Learning Integrated Approach for the Investigation of Honey Adulteration

In this study, the efficacy of multi-modal data fusion in detecting honey adulteration using spectral data from FT-NIR, FT-MIR, and CRM sensors was explored. Feature-level fusion was implemented to develop two distinct scenarios for assessing the performance of three one-class classifiers: OC-SVM, LOF, and Isolation Forests. The results indicate that the fusion approach significantly enhances the detection accuracy of adulterated honey samples. Notably, the OC-SVM classifier demonstrated superior performance in scenarios where data from multiple spectral instruments were integrated, achieving high recall rates and effectively reducing false positives. Indeed, it achieved a recall rate of 95%, precision of 92%, and an F1-score of 93.5%. The high performance indicates the potential of the use of non-invasive spectral techniques in honey adulteration.