Fish oil is becoming increasingly popular as a dietary supplement as well as for its use in animal feed, which is mainly due to its high contents of the health promoting omega-3 fatty acids. However, these polyunsaturated fatty acids are highly susceptible to oxidation, which results in a decrease of the fish oil quality. This study investigated the potential of 1H NMR, FT-MIR, and FT-NIR spectroscopy in the quality assessment of fish oils. A total of 84 different fish oils, of which 22 were subjected to accelerated storage with varying temperature and light exposure, were used to develop models for predicting the peroxide value (PV), the anisidine value (AnV), and the acid value (AV). Predictions were based on comprehensive spectroscopic data in combination with Artificial Neural Networks (ANN) as well as Partial Least Squares Regression (PLSR). The best ANN model for PV was obtained from NMR data, with a predictive coefficient of determination (Q2) of 0.961 and a Root Mean Square Error of Prediction (RMSEP) of 1.5meqO2kg-1. The combined MIR/NIR data provided the most reliable ANN model for AnV (Q2=0.993; RMSEP=0.74). For AV, the ANN model based on the MIR data yielded a Q2 of 0.988 and an RMSEP of 0.43mgNaOHg-1. In most cases, the accuracy of the ANN models was superior to the respective PLSR models. Variable selection and data dimensionality reduction turned out to improve the performance of the ANN models in some cases. The application of 1H NMR, FT-MIR, and FT-NIR spectroscopy in combination with ANN can be considered very promising for a rapid, reliable, and sustainable assessment of fish oil quality.
Keywords: Anisidine value; Artificial neural networks; Data fusion; Infrared spectroscopy; Nuclear magnetic resonance spectroscopy; Peroxide value.
Copyright © 2017 Elsevier Ltd. All rights reserved.