Predicting Cattle Meat Types through Machine Learning Models Trained on Rapid Evaporative Ionization Mass Spectrometry (REIMS) Data
Título de la revista
ISSN de la revista
Título del volumen
Zamorano: Escuela Agrícola Panamericana
Currently, the meat industry faces several problems, one of them being consumer fraud, which has arised the need to guarantee to the consumer that the product offered for sale is what it says on the label. This leads to the main objective of this study, which was to use data obtained by Rapid Evaporative Ionization Mass Spectrometry (REIMS) analysis for the training of 12 predictive models in three different dimensional reduction methods in order to train these models for the quick and accurate identification of the bovine breed from which the meat is obtained. Steaks from the Loggisimus dorsi muscle in the USDA classification as "Prime" from the Angus breed and the Wagyu breed were used. Each method's five best predictive models were selected for analysis and discussion. The best dimensional reduction method was Feature Selection (FS), which showed accuracies ranging from 73.6 to 91.8% in the different predictive models, being the best predictive model SVM Poly, which obtained the highest percentages in the performance metrics in the three dimensional reduction methods. Thus, demonstrating the effectiveness of using REIMS data for predicting the bovine breed from which the Longissimus dorsi steaks derive.
Dimensional reduction, Feature selection (FS), predictive model, prime, Wagyu