An Instrumented Approach to the Classification of Alcoholic Beverages: Combining PTR-ToF Mass Spectrometry, with Machine Learning

Herbert, James Feleciano (2018) An Instrumented Approach to the Classification of Alcoholic Beverages: Combining PTR-ToF Mass Spectrometry, with Machine Learning. [USQ Project]


The human sense of smell is incredibly complex. It has the power to discriminate between even the smallest of molecular changes within an odour (Buck, 2005). The primary source of an odour, although certainly not solely, are ‘Volatile Organic Compounds’, or ‘VOCs’. In the field of Food Science there is great interest in the relationship between Volatile Organic Compounds, and the quality, or condition, of the substance(United States Environmental Protection Agency, 2017). This study will attempt to establish a method of analysis, of the chemical composition of human consumables. The primary aim, to establish an alternative method of chemical analysis, avoiding costly, and time-consuming processes.

Machine Olfaction, the ability for artificial devices to mimic Olfaction is no longer considered as a novel concept. Mass Spectrometers, although costly are an incredibly reliable means of implementing Machine Olfaction. They can be quite complex to utilise, however, this complexity allows them to be more robust. Machine Olfaction systems are also quite commonly coupled with Statistical, and Machine Learning algorithms. Odorant Data is often quite 'wide'. 'Wide' data sets consist of many independent variables, also known as features. ‘Principle Component Analysis’, or ‘PCA’, is used to identify the features that provide the most influence (Yan et al., 2017).

The chosen target substance for analysis will be Beer. Beer is widely available, and has a wide range of aromas, and flavours. Twelve samples, of varying types of Lagers, Ales, and Stouts were analysed using a ‘Proton Transfer Reaction Time-of-Flight Mass Spectrometer’, or ‘PTR-ToF MS’. Once this data was acquired with Mass Spectrometer, Principle Component Analysis was applied to the data set. The resultant data from the PCA was used to train a ‘Tree Bagging’ classificati0n model, a type of ‘Bagged’ Decision Tree . The model was trained to discriminate the samples by its Style and Type; i.e., whether it was a Pale Lager, Dark Ale, Wheat Beer, etc. Training of the Classification Model was performed using two-thirds of the data set; the remaining one-third was used for validating the Model.

Unfortunately, the Classification Model was unable to reliably classify the validation data set. Although the outcomes of this study were not ideal, there still exists potential with future work. In this study, the entire chemical profile of a substance was used to form a correlation with its Style, or Type. It may be more reliable to look at particular components of its chemical profile, and identify relationships with certain dietary components present within the sample. For example, Alcohol content, Hops content, Calorific content, etc.

Statistics for USQ ePrint 40702
Statistics for this ePrint Item
Item Type: USQ Project
Item Status: Live Archive
Faculty/School / Institute/Centre: Historic - Faculty of Health, Engineering and Sciences - School of Agricultural, Computational and Environmental Sciences (1 Jul 2013 - 5 Sep 2019)
Supervisors: Lobsey, Craig; Brett, Peter; Dunlop, Mark
Qualification: Bachelor of Engineering (Honours) (Computer Systems)
Date Deposited: 31 Aug 2022 01:19
Last Modified: 29 Jun 2023 01:40
Uncontrolled Keywords: Machine Olfaction; Mass Spectrometers

Actions (login required)

View Item Archive Repository Staff Only