Abstract

Cattle with age under 30 months (UTM) are usually slaughtered for human consumption to be not associated of the disease called mad cow. Though the age of the animal can generally be estimated by dentition, this method cannot be applied to a piece of meat from which the teeth have been removed. Since volatile organic compounds have been used to analyze food samples, this technique was used to recognize meat obtained from cattle aged over 30 months (OTM). The monitoring of the volatile organic compounds (VOCs) released by UTM and OTM beef were done by using gas chromatography. A dataset with more than 500 chromatograms (each one with 17 VOCs fully identified) from fresh meat, refrigerated meat and vacuum-packaged meat was used to develop a classifier by using neural networks. Neural networks were trained with backpropagation, and then further optimized by using meta-learners. Optimal configuration of the neural networks allowed discriminating between beef obtained from OTM or UTM cattle with accuracy close to 90 %. Results contrast favorably with more traditional statistical methods such as linear discriminant analysis (LDA), soft independent modeling of class analogies (SIMCA), partial least square discriminate analysis (PLS-DA), and support vector machines (SVM). In conclusion, VOCs can be used as a fingerprint to recognize OTM beef from a pool of meat obtained indistinctly from fresh meat, refrigerated, or vacuum-packaged meat.