Effects of Pre-Processing and Principal Components for Artificial Neural Network in Non-Destructive Internal Quality Prediction of Mango across Different Harvest Periods

Abstract

Dry Matter content (DMC) is one of the important components can be used to determine the quality of mango. However, DMC measurement is destructive and tedious. Although near infrared spectroscopy (NIRS) is a promising fast and non-destructive analytical technique, the accuracy of NIRS is susceptible with the use of pre-processing methods throughout the NIRS modeling process across different harvest periods. Thus, this research aims to investigate the effects of pre-processing methods on two popular different machine learning models (i.e. artificial neural network (ANN) and principal components ANN (PCs-ANN) in non-destructive internal quality prediction using NIRS. Two different pre-processing methods i.e. Standard Normalized Vector (SNV) and second order Savitzky-Golay (SG) would be applied to pre-process the near infrared spectra prior to the machine learning process. Result indicates that both ANN and PCs-ANN with second order SG outperformed that with SNV and that without pre-processing method. Additionally, based on root mean squared errors of validation (RMSEV), ANN outperformed PCs-ANN; while PCs-ANN outperformed ANN when root mean squared errors of prediction (RMSEP) was referred to. This indicates that ANN tends to be over-fitted while the involvement of principal component improves the robustness of ANN when it is applied to new samples that harvested from different periods.