RECURSIVE ALGORITHM FOR PARTIAL LEAST-SQUARES REGRESSION

In this paper an algorithm is presented for updating partial least squares (PLS) regression models with new calibration objects. The new data are included in the model by recursive updating of the loading vectors. This is done by representing old data by modifications of their loading matrices, while new data are represented by their x and y vectors. The algorithm keeps the sizes of the matrices that go into the PLS algorithm constant, instead of using ever growing X and Y matrices. The algorithm will be of use in any situation where new data are to be included in the calibration model after the initial calibration. Simulations have been run where the recursive algorithm is compared with traditional PLS.