Introduction of LSSVR for the Prediction of the Yellowness Index

Abstract

Data-driven models including principal component regression (PCR), partial least square regression (PLSR), and least square support vector regression (LSSVR) have been widely applied as predictive models in various applications. However, studies employing regression models to estimate the yellowness index (YI) are scarce in the literature. This study, therefore, focuses on developing non-destructive YI measurements using regression models. The collected RGB calculated XYZ and obtained CIE LAB values were set as the input variables. Meanwhile, the YI value was denoted as the output variable. Results indicated that the LSSVR model outperforms PCR and PLSR models in predicting YI in which the root means square errors of LSSVR for the training and testing datasets were found to be 261,406% to 294,218% and 725% to 772% lower than PLSR and PCR, respectively. LSSVR is also attributed to higher coefficients of determination (R 2 ) that are superior to PLSR and PCR, whose R 2 values are very close to 1. Nonetheless, the computational times of training and testing datasets for LSSVR are much longer than those of PLSR and PCR. Consequently, a highly accurate LSSVR model-based YI sensor shows promising applications particularly if the computational load can be further minimized.