Information Feature Compression Based on Partial Least Squares

Partial least squares (PLS) regression is introduced for information feature compression, which is proven to be more advantageous than the approach of principal component analysis (PCA) in its simplicity,robustness, and clearness of qualitative explanation. It is powerful for processing multicollinear information, particularly when the number of predictor variables is large and the sample size is small. Meanwhile it is more effective than PCA, when the information of data matrix X is compressed while still maintaining their maximum correlation with objective matrix Y. Numerical example shows that the algorithm is feasible and effective, and provides a new research approach for information feature compression in pattern recognition.