A Supervised PLS Information Recognition Algorithm

Abstract

This paper analyzes the basic principle of Partial Least Squares Regression (PLS), and based on PLS, a Supervised Information Recognition (SIR) algorithm is set up. The algorithm is built up by regarding response variables as 0-1 variables, and it organically combines feature extraction and classifier design in pattern recognition. Compared with Fisher discriminative analysis, Bayes discriminative analysis and some other classical pattern recognition algorithms, SIR algorithm has a much more powerful information recognition ability, and it doesn't demand the data to have good distribution, especially it is more effective to collinearity information or explaining variables are more but the sample size is small. Applied this algorithm to classification recognition of land quality, the results show that the algorithm established in this paper is effective and reliable.