Robust DEEP heterogeneous ensemble and META-learning for honey authentication.

Food fraud raises significant concerns to consumer health and economic integrity, with the adulteration of honey by sugary syrups representing one of the most prevalent forms of economically motivated adulteration. This study presents a novel framework that combines data from multiple analytical techniques with specialized deep learning models (convolutional neural networks), integrated via meta-learning, in order to differentiate between pure honey and samples adulterated with sugar cane molasses, glucose syrup, or caramel-flavored ice cream topping. Unlike traditional chemometric methods, this approach expands the input feature space, leading to enhanced predictive performance. The resulting deep heterogeneous ensemble learner exhibited considerable generalization capability, achieving an average classification accuracy of 98.53 % and a Matthews correlation coefficient of 0.9710. Furthermore, the ensemble demonstrated exceptional robustness, maintaining an accuracy of 73 %, even when 90 % of the input data were corrupted, underscoring its unparalleled capacity to generalize under both subtle and extreme data variability. This adaptable and scalable solution underscores the transformative potential of ensemble-meta-learning strategy for addressing complex challenges in analytical chemistry. The model, its constituents and other additional resources were made available in an open repository.