A simple idea on applying large regression coefficient to improve the genetic algorithm-PLS for variable selection in multivariate calibration

• A new and simple idea that applies PLS large regression coefficient to improve GA-PLS.
• It not only has a better starting and ending response, but also obeys the rule of the GAs.
• There is much improvement compared to the original GA-PLS.

Genetic algorithm-based couple with partial least squares (PLS) has been successfully applied for variable selection in multivariate calibration. On the basis of the fact that a large PLS regression coefficient indicates an important variable, a new and simple idea that the structure of a proportion of chromosomes in the initial population is determined by the large regression coefficient is presented in this study. The regression coefficient is obtained by establishing the PLS modeling on the autoscaled data. With this improved approach, the modified GA-PLS method not only makes the optimization better toward the optimal solution, but also obeys the rule of the GAs. The results obtained through investigating one simulated dataset and two near infrared dataset show that the modified method has made much improvement on variable selection compared to the original GA-PLS.