Chemometric pre-processing can negatively affect the performance of near-infrared spectroscopy models for fruit quality prediction

Chemometrics pre-processing of spectral data is widely performed to enhance the predictive performance near-infrared (NIR) models related fresh fruit quality. Pre-processing approaches in domain NIR analysis are used remove scattering effects, thus, enhancing absorption components chemical properties. However, case fruit, both and properties key interest as they jointly explain physicochemical state a fruit. Therefore, that reduces information spectra may lead poorly performing models. The objectives this study test two hypotheses explore effect on first hypothesis with scatter correction techniques can reduce useful correlated property interest. second Deep Learning (DL) model raw absorbance (mix absorption) much more efficiently than Partial Least Squares (PLS) regression analysis. To hypotheses, real set dry matter (DM) prediction mango was used. dataset consisted total 11,420 reference DM measurements for training independent testing. chemometric methods explored were standard normal variate (SNV), variable sorting normalization (VSN), Savitzky-Golay based 2nd derivative their combinations. Further modelling i.e., PLS DL evaluate pre-processing. results showed best root mean squared error (RMSEP) obtained data. general decreased Further, attained lowest RMSEP 0.76%, which 13% lower compared should be carefully while analysing