Decision trees as feature selection methods to characterize the novice panel’s perception of Pinot noir wines

Abstract Consumer perceptions and purchase behaviours are typically determined by phenolic influenced sensory attributes such as astringency, bitterness, and sourness. Marketing strategies would be more effective if machine learning methods assisted winemakers in understanding the chemical parameters that influence panellists' sensory evaluations of Pinot noir wines. Nowadays, feature selection methods such as random forest classifier and neighbourhood component analysis are utilised to select important factors. In the meantime, decision trees are utilised in regression or classification models as opposed to feature selection methods. In this study, decision trees were able to identify the relationships between sensory attributes and important chemical parameters in Pinot noir wines from diverse product extrinsic cues (regions of origin, vintages and price points) and novice panels. Decision trees not only could be utilised to build soft sensors but also work as effective feature selection methods, which could inspire winemakers to make proper Pinot noir wines. With the help of principal component analysis and scatter plots, it was discovered, for instance, that total phenolics, total flavanols, total flavan-3-ols, and total tannins in wines could considerably contribute to astringency, bitterness, and sourness and that total anthocyanin could contribute to clarity regardless of diverse Pinot noir wine or novice panels.