Conclusion
In this study, we have shown the feasibility of applying NIR
spectrometry for the determination of the fatty acid composition
in sea buckthorn seed oil for the
first time. The number of variables
selected by the UVE-PLS, CARS-PLS and UVE-SPA-MLR models can
be reduced without compromising the prediction performance
compared with the PLS model with full spectrum. This suggests
greater stability for the calibration models. We also illustrated that
the application of variable selection methods could make the
models more reliable and simpler. The performance of the UVESPA-
MLR model based on a small subset of variables showed that
UVE-SPA is an effective approach for the selection of variables
using the MLR model as a straightforward calibration method with
greater accuracy than obtained using the the well-established PLS
method only.