Prediction of pig fatty acids composition by near-infrared spectroscopy using neural networks and support vector machine [post]

NING PAN, chenglong TAO
2022 unpublished
This work reports on the development and optimization of NIRs technology for the prediction of fatty acids in pig carcasses in the slaughterhouse; use of this technology would enable implementation of a data pretreatment and modelling system. Two years of spectral data from different producers were used to construct a robust model that can predict four fatty acids. For outlier detection of high-dimensional data, an optimized method, the optimized Local Outlier Factor (LOF), was used to remove
more » ... tliers. Two correction operations and a convolution operation were combined into different data processing methods and were compared according to the prediction accuracy. Models were developed for the prediction of the four fatty acids in Iberian pig fat, including Principal component neural networks(PCNN), Back propagation neural network(BP neural network), Least Square Support Vector Machine(LS-SVM), and 45 folds Least Square Support Vector Machine(45 folds LS-SVM). The processed data can obtain the smallest prediction error by developing LS-SVM model, obtaining for palmitic acid Root Mean Square Error of Prediction (RMSEP) values of 0.237, for stearic acid 0.247, for oleic acid 0.413 and for linoleic 0.113, respectively. These results confirm the optimization effect of appropriate data pretreatment and modelling system on the prediction of fatty acids.
doi:10.21203/rs.3.rs-1811254/v1 fatcat:kctcs4saafds7cdcqbcjig3ysy