Penerapan Regresi Komponen Utama Kekar dan Regresi Kuadrat Terkecil Parsial Kekar dalam Pemodelan Kalibrasi Multirespon Kayu Jati

Abstract

Calibration modeling is a method which often be used to estimate chemical contents of a material from measured spectra. The problems in calibration modeling are the number of independent variables larger than the number of observations, multicollinearity between independent variables, and outliers. RPCR and RSIMPLS are robust methods based on PCR (Principal Component Regression) and PLS (Partial Least Square) algorithms capable to solve those problems. A modified method of RPCR and RSIMPLS, called RPCR-M and RSIMPLS-M. They are alternative methods used weight in RPCR and RSIMPLS based on M-estimators with Huber weight function. Both modified methods are applied to multirespon data to estimate chemical contents of teak wood and compared. The results of validation and simulation showed that RPCR-M is better when the number of extreme outliers are less then two, whereas RSIMPLS-M is better and more stable when extreme outliers in the data are more then two.