New approach for barrel-aged distillates classification based on maturation level and machine learning: A study of cachaça

Abstract

The evolution of distillate beverages during the aging process in wooden barrels is related to the extraction and transformation of wood compounds. Even though time is an important factor for the quality of distillates, other attributes concerning wooden barrel features such as wood species, internal toast, and previous use are of paramount importance to confer maturation characteristics to the beverage. We evaluated the level of lignin-derived products (LDPs) in Brazilian sugar cane spirit (cachaça) aged in different barrels using high performance liquid chromatography with ultraviolet/visible (HPLC UV–Vis) analysis, and compared the results to aging classes reported for distillates applying multivariate principal component analysis (PCA). Unsupervised machine learning techniques of clustering were applied to identify similarities among samples and form cohesive groups concerning wood derived phenolic compounds. In an effort to provide a new metric for distillates aged in wooden barrels, four new classes of maturation were determined as a qualitative parameter. We describe their characteristics, based on quantitative parameters equivalent to guaiacyl and syringyl compounds, as well as a regression model for the classification of new samples (86.9% of prediction accuracy).