Partial Least Squares (PLS) regression is of one of two multivariate data analysis model types available in Evince. PLS is useful when it is desirable to obtain a prediction model that can foretell a number of properties from some kind of data. For example, a PLS model can be constructed to predict the price of different cars from their physical specifications.

PLS uses two data matrices for the regression, X and Y. The PLS algorithm finds components in X that explains as much as possible of the variation in Y. For each PLS component, a vector w is calculated, which gives the maximum covariance between X and Y. The model between X and Y can be written:

Y = XB + F

where F is a residual matrix. PLS regression coefficients, B, are generated as

B = W(P'W)^-1C'

where C is the loading matrix in the Y matrix.