Understanding of the metabolic fingerprints of lamb meat (via Rapid Evaporative Ionisation Mass Spectrometry REIMS) associated with pH via WEKA machine learning platform

Abstract

The ultimate pH of a muscle is closely related to meat quality. It is highly dependent on the types of muscle fibres where intermuscular difference between fast glycolytic and slow oxidative muscles affects post-mortem metabolism. The aim of the current study was the prediction of ultimate pH via WEKA machine learning by using the Rapid Evaporative Ionisation Mass Spectrometry (REIMS) features to evaluate the classification of data with the interference of stressors. Ten different muscles from 20 lamb carcasses (10 control group, no exercise while another 10 from treated group, exercise) were subjected to REIMS analysis at slaughter and pH measurement at 24h post-mortem. Data from these measurements was preprocessed into datasets with three selection algorithms; feature selection, clustering (via simpleKmeans) and feature extraction (via Principal Component), generated three sets of reduced datasets (smaller number of attributes). The predictive accuracies of the three machine learning algorithms/ classifiers (Simple Linear Regression, Multilayer Perceptron and Random Forest) of each individual reduced dataset were compared based on the Mean Absolute Error (MAE). The impact of muscle attribute on the predictive accuracy was also examined using the same algorithms. Overall, Random Forest classifier model using feature selection approach had the lowest MAE value, 0.59 and 0.581 for “without muscle attribute” and “with muscle attribute”, respectively. The low MAE value indicated the higher predictivity of the model. The presence of muscle attribute lowered the MAE value, showing the correlation between ultimate pH and types of muscles.