Detecting Forged Alcohol Non-invasively Through Vibrational Spectroscopy and Machine Learning

Alcoholic spirits are a common target for counterfeiting and adulteration, with potential costs to public health, the taxpayer and brand integrity. Current methods to authenticate spirits include examinations of superficial appearance and consistency, or require the tester to open the bottle and remove a sample. The former is inexact, while the latter is not suitable for widespread screening or for high-value spirits, which lose value once opened. We study whether non-invasive near infrared spectroscopy, in combination with traditional and time series classification methods, can correctly classify the alcohol content (a key factor in determining authenticity) of synthesised spirits sealed in real bottles. Such an experimental setup could allow for a portable, cheap to operate, and fast authentication device. We find that ethanol content can be classified with high accuracy, however methanol content proved difficult with the algorithms evaluated.