A handheld near infrared (NIR) spectrometer was used for on-site determination of the fatty acids (FAs) composition of industrial fish oils from fish by-products. Partial least square regression (PLSR) models were developed to correlate NIR spectra with the percentage of saturated fatty acids (SFAs), monounsaturated fatty acids (MUFAs), polyunsaturated fatty acids (PUFAs) and, among them, omega-3 (ω-3) and omega-6 (ω-6) FAs. In a first step, the data were divided into calibration validation datasets, obtaining good results regarding R2 values, root mean square error of prediction (RMSEP) and bias. In a second step, all these data were used to create a new calibration, which was uploaded to the handheld device and tested with an external validation set in real time. Evaluation of the external test set for SFAs, MUFAs, PUFAs and ω-3 models showed promising results, with R2 values of 0.98, 0.97, 0.97 and 0.99; RMSEP (%) of 0.94, 1.71, 1.11 and 0.98; and bias (%) values of -0.78, -0.12, -0.80 and -0.67, respectively. However, although ω-6 models achieved a good R2 value (0.95), the obtained RMSEP was considered high (2.08%), and the bias was not acceptable (-1.76%). This was corrected by applying bias and slope correction (BSC), obtaining acceptable values of R2 (0.95), RMSEP (1.09%) and bias (-0.05%). This work goes a step further in the technology readiness level (TRL) of handheld NIR sensor solutions for the fish by-product recovery industry.
Keywords: chemometrics; circular economy; fish oil industry; lipid profile; no-waste; omega-3; recovery; reuse; smart sensors.