Ridge regression for NIR analysis with multicollinearity Academic Article uri icon

abstract

  • High intercorrelation between absorbances at different wavelengths is common in near infrared (NIR) analysis. NIR reflectance analysis was conducted to predict carotene in fresh tomatoes. When linear regression is employed the estimated parameters are practically random numbers, however high correlations are obtained between the predicted and true values (R=0.78). Ridge regression yields estimators with normal values, with lower parameter correlations (R=0.74). However, ridge regression is capable of overcoming noise versus linear regression which is not capable of predicting carotene in the presence of minor noise and multicollinearity.

publication date

  • January 1, 2001