Recent interest in the antioxidant capacity of foods and beverages is based on the established medical knowledge that antioxidants play an essential role in counteracting the damaging effects of free radicals, preventing human neurodegenerative diseases, cardiovascular disorders, and even cancer. At the same time, there is no "the method" that uniquely defines the antioxidant capacity of substances; moreover, the question of interrelation between results obtained by different experimental techniques is still open. In this work, we consider the trolox equivalent antioxidant capacity (TEAC) values obtained by electron paramagnetic resonance (EPR) spectroscopy and ultraviolet-visible (UV-vis) spectroscopy using the classic objects for such studies as an example: red, rosé, and white wine samples. Based on entirely different physical principles, these two methods give values that are not so simply interrelated; this creates a demand for machine learning as a suitable tool for revealing quantitative correspondence between them. The consideration consists of an approximate correlation-based analytical model for the key argument (i.e., TEACEPR) with subsequent adjustment by machine learning-based processing utilizing the CatBoost algorithm with the usage of auxiliary chemical data, such as the total phenolic content and color index, which cannot be accurately described by analytical expressions.
Keywords: CatBoost machine learning; EPR spectroscopy; UV–vis spectrophotometry; antioxidant capacity; wine.