Overcoming multicollinearity in near infrared analysis for lycopene content estimation in tomatoes by using ridge regression

High intercorrelation between absorbance at different wavelengths is common in near infrared analysis and was observed in an experiment to determine l ycopene in tomatoes. Simulation analysis and experiments were conducted to estimate the effects of this problem on the estimators and on the predictiv e ability of linear regression and ridge regression. Applying linear regres sion to the experimental data resulted in very large parameter values, impl ying poor predictive ability. When linear regression gives very large param eter values, the estimated parameters are practically random numbers and ar e not correlated to the true ones. Ridge regression yielded estimators with normal values, but which are still poorly correlated with the true paramet ers. However, the predictive ability of the derived equation is good and ma y be used in practice to determine lycopene content in tomatoes since it is relatively easy to update.