ABSTRACT
The Kennard-Stone algorithm was used to select Eucalyptus spp. wood samples for development of NIRS (Near-Infrared Spectroscopy) calibration models aiming to minimize number of samples but maintaining the model precisions. A large number of Eucalyptus spp. wood samples (3369 samples) were used to develop NIRS calibration models for the wood basic density, the lignin content and the ethanol-toluene extractives. The models developed with the total number of samples were compared with models developed using only 1000, 500, 200 and 100 samples, which were selected using the Kennard-Stone algorithm. Analysis of the models statistics parameters confirmed the similarity of all models, with exception of the 100 sample models, demonstrating the possibility of substantial savings in time and costs for wood laboratory analysis.
Keywords:
algorithm; prediction; sampling; wood quality