Machine learning in whisky identification and verification

We worked with the Scotch Whisky Research Institute (SWRI) to investigate how data analytics could be used to tackle counterfeiting across the sector.

Challenge

The SWRI carries out pre-competitive fundamental research on behalf of its members, representing approximately 90% of the production capacity of the sector. SWRI offers analytical services, using traditional techniques such as GC-FID (Gas Chromatography – Flame Ionisation Detector) or GC-MS (Gas chromatography–Mass Spectrometry) to detect counterfeits. Using these methods to distinguish between different samples can be challenging, given the complexity resulting from the number of different compounds present in the vapour above a whisky sample. SWRI wanted to explore analytics and AI models to improve the identification of authentic Scotch whisky using traditional GC-MS data and more complex GCxGC-MS sampling, as developed at The Open University.​

Approach

The team developed machine learning models to identify spectral regions of relevance, distinguish between real and counterfeit whiskies, and determine the likelihood a given sample is from a previously identified whisky. Models were applied to both one-dimensional GC-MS data and two-dimensional GCxGC datasets, and developed into an application that enabled visualisation of the spectra of a whisky sample in the context of previous known samples. The tool predicts the veracity of a given sample corresponding to a genuine Scotch whisky.​

Benefits

As a premium product, Scotch whisky is a counterfeiting target. Identifying more sophisticated counterfeit samples requires more sensitive analytic techniques providing analyte-rich whisky profiles and identify unknown markers for counterfeit detection. This is challenging for existing analytical methodology and the task of automating processing of large data files, ultimately becoming prohibitive to implementing routine authentication provisions due to cost implications. The SWRI looks forward to exploring these challenges further as part of the Innovation Return on Research (IROR) programme, a collaboration between STFC and IBM Research.​​

“Without such efficient data processing techniques, the time cost prevents such techniques ever​​ becoming a part of a routine authentication provision within the Scotch Whisky sector. We look forward to exploring ​​the potential of the ​advanced analytical and data processing techniques developed in conjunction with the Open University, IBM and the STFC Hartree Centre to help us combat counterfeiting​. Scientific progress is ​most effective when different ​ideas and expertise can collaborate.”

Ian Goodall, Scotch Whisky Research Institute