Geo-tracing of Coffee Beans Using Metal Oxide Semiconductor Gas Sensors Array

The determination of food freshness and quality using human senses can be biased. Coffee is a beverage known for its popularity and increase in demand. The quality and price of coffee differ based on its geo-origins. This paper reports on the application of metal oxide semiconductor (MOS) gas sensors array in geo-tracing of coffee beans. The MOS gas sensors array, consisting of four models of sensors, was used for headspace volatile organic compound (VOC) sampling of coffee beans from Brazil, Ethiopia, Columbia and Vietnam to generate sensor responses profile corresponding to each geo-origin. Principal component analysis (PCA) was employed on the sensor responses data for dimension reduction, followed by multiclass classification to geo-trace each coffee bean sample. The methodology yielded an overall coffee beans geo-tracing accuracy up to 97.5% using classification models such as fine gaussian support vector machine (SVM), fine k-nearest neighbour (KNN), weighted KNN, boosted trees, bagged trees, subspace KNN and random under-sampling (RUS) boosted trees.