Beef spoilage assessment using e-nose and machine learning on unbalanced dataset

Abstract

Food quality and freshness especially meat which have short shelf life like beef meat is a real problem in nowadays. This kind of food should be stored at suitable temperature and humidity conditions. For this purpose, a system is created to detect different states of freshness through an open unbalanced dataset. Machine learning models' performance is affected by unbalanced classes, which leads to biased outcomes and poor performance on minority classes, to address this issue this study uses synthetic minority oversampling technique (SMOTE). For this purpose, an open dataset containing 10800 samples, where four classes are distinguished (excellent, good, acceptable and spoiled). In this study, the proposed e-nose is composed of 10 sensors. For classification 6 machine learning methods are used. The best results are obtained from k-nearest neighbors (KNN) model with 99.83% of accuracy, 99.86% of precision, 99.80% of recall and 99.83% of F1-score.