The Accuracy of near Infrared Prediction of Hemicellulose Content Arising from Varying Introduced Errors

Abstract

Reference method noise is one of the important factors which affects the accuracy and the precision of NIR predicted values. In this paper, noise was deliberately and artificially added to the reference data of hemicelluloses content of Acacia spp. in four different ways, namely adding absolute error, relative error, random absolute error or random relative error. The effect of the addition of different error to the reference data on NIR calibration models and their prediction results were studied. Although the results of calibration models were very poor when different errors were added to the reference data, using the resulting model to predict values for unknown samples to which errors were not added, resulted in predictions that were better than expected, especially for addition of random absolute error and random relative error condition. The results indicate that the NIR calibration models produce predicted values that are acceptable if the noise is not too large.