Some Methods of Variable Selection

Implementing genetic competition requires definition of a fitness function for the chromosome population, e.g., (correlation coefficient) or / cv (correlation coefficient for CV). 

Leardi and Gonzalez have addressed the issue of the use of GAs for extraction of the most relevant variables for PLS analysis. The critical point is summarized in their paper ... a variable/objects ratio equal to 5 has been found to be the critical point, beyond which using GA will be very dangerous [27]. 

The Moby digs software of Todeschini [28] applies the GAfor variable selection in a MLR algorithm. The variable pool consists of 2000 variables at maximum two populations are allowed at a time. As a fitness function, can be selected. 

The pairwise correlation method (PCM) [29,30] utilizes a portion of information present in the data but overlooked till now. PCM selects from two independent variables (Zj and Zj) which one is superior, i.e., correlates better to the dependent variable T. Three vectors are defined Y (dependent variable), Zj, and Zj (independent variables). The task is to choose the superior one from Zj and Zj. First, it is assumed that both of the independent variables correlate positively with the dependent variable Y. Other cases are discussed in refs. [29,30] exhaustively. All the possible element pairs of the Y vector are considered that can occur when the differences AZj for Y vs. Zj, and AZ2 for Y vs. Z2 are determined. Only the signs of 

In its generalized form (GPCM), all possible independent variable pairs are compared and the number of superiority is determined. The number of superiority  

---