Quality Classification of Foods Through Analysis by Machine Learning

Introduction

Foods are comprised of numerous components and the quality of foods may not be completely the same even for the same food products. Differences in quality are considered to be caused by slight differences in the components that comprise food products. Therefore, for the purpose of a complete quality evaluation, comprehensive analysis of components is gaining attention in recent years. In order to estimate and identify the subjective properties of foods such as taste, smell, and deterioration based on their components, one method that is expected to be effective is to learn the relation between components and subjective properties of a known sample and then utilize those results for an unknown sample. This article studies whether or not it is possible to distinguish between beef samples that have been properly refrigerated and those that are expected to have some deterioration from being exposed to a 40 °C environment for 3 hours based on the analysis results of volatilized components when those samples are heated to 200 °C. After making a classifier learn known data of each sample type, the learned data was used to define quality. This was then used to classify unknown data as either sample type and calculate the percentage of correct results. By using a support vector machine (SVM) as the classifier, we were able to obtain correct results by 95.8 % even for samples that were hard to classify by comparing chromatograms or through principal component analysis of peak area values.

March 26, 2019 GMT