Corrects the prior all-malt dataset and chemical/production-input claims. Scotch training products include blends, malts and grains.
Sensory profiles are the input and Scotch/non-Scotch is the target; this is not chemical-to-sensory regression.
The neural network used mean expert-panel ratings on thirteen aroma attributes from 144 training products: 72 Scotch and 72 non-Scotch. A separate twelve-product prediction set assessed category classification.
Table 3 and Sections 2.1–2.5
Printed pp. 164–166 / PDF sheets 2–4
Training and prediction samples; neural-network methods
Corrected against all ten pages, three tables and twelve figures in September 27, 2026 audit; earlier paraphrase misrepresented this study. Original record and relations preserved.
Whiskey Knowledge Databases › Evidence Excerpts
Verified Paraphrase checked against the complete supplied source.
Evidence
Sensory profiles from 144 Scotch malt whisky spirits were used to test statistical models intended to connect chemical or production information with perceived character.
Locator and context
Locator: Page 1
Source section: Abstract
Context: A large expert-profile dataset underlies the modelling exercise.
Interpretation and caution
Dataset scale is strong, but generalization beyond Scotch malt spirits requires caution.
Verification
Paraphrase and locator checked against the complete locally preserved source during the 2026-09-05 full-source pass.