Question

Why is partial least square good with missing data?

Answer 1

Why Partial Least Squares (PLS) algorithm is good with missing data:

PLS is a successful tool in chemometrics. The PLS algorithm is used to solve the latent variables, a kind of missing data problem that cannot be observed directly. Several kinds of PLS algorithms can be widely used for estimating the value of latent variables. One approach is combining traditional linear regression type PLS algorithms with missing data handling methods. Also, it introduces quantile regression improving the performances of PLS algorithms when the relationships among manifest and latent variables are not fixed per the explored quantile of interest. PLS algorithms perform well when missing manifest variables occur. PLS algorithm is a new algorithm that is used for dealing with missing values in predictive modeling. A trimmed score regression (TSR) adaptation is proposed from PLS model exploitation with missing values. PPLS is used for building the multivariate calibration models. The PLS algorithm appraises incomplete data. The PLS algorithm is becoming popular in analyzing interactive marketing applications which may be attributed to its robustness and accuracy when data are abnormally distributed. PLS provides a data classification mechanism with missing data handling.

In the iterative replacement method:
The PLS algorithm consists of the following steps:
1) It replaces missing data by an initial value.
2) It optionally centers and scales the matrices.
3) From the resulting data matrices, it calculates the PLS model.
4) It obtains new values for the missing variables or data as the corresponding entries in the prediction matrices.
5) In case, convergence is not reached, it returns to step (2).

PLS carries out the critical step in its algorithm which is to calculate the imputations for the missing measurements.