Can Data Science Predict Human Perception of Gin?

Sensory perception involves processing a wide array of stimuli from the environment. In terms of assessing quality of gin, the stimuli include a wide array of compounds derived from the botanicals used in production. In this work, the composition of 175 gin samples was determined using qualitative methods based on Gas Chromatography Mass Spectroscopy with Solid Phase Micro Extraction (GCMS SPME). A total of 221 compounds were detected. Sensory data on the same gin samples was collated from industry award panels and used to establish a link between high scores and the presence or absence of the 221 compounds identified. Through a machine learning tool, patterns of both negative (15 compounds) and positive (9 compounds) factors were attributed to sensory score by the panellists. The model developed was able to predict sensory score with an accuracy of over 90%, making it a useful tool to augment or validate sensory panel scores and could even lay the foundations for legislative frameworks for the categorisation around what flavours makes gin.