Determining the number of components in PLS regression on incomplete data set

Abstract

Partial least squares regression – or PLS regression – is a multivariate method in which the model parameters are estimated using either the SIMPLS or NIPALS algorithm. PLS regression has been extensively used in applied research because of its effectiveness in analyzing relationships between an outcome and one or several components. Note that the NIPALS algorithm can provide estimates parameters on incomplete data. The selection of the number of components used to build a representative model in PLS regression is a central issue. However, how to deal with missing data when using PLS regression remains a matter of debate. Several approaches have been proposed in the literature, including the Q2 criterion, and the AIC and BIC criteria. Here we study the behavior of the NIPALS algorithm when used to fit a PLS regression for various proportions of missing data and different types of missingness. We compare criteria to select the number of components for a PLS regression on incomplete data set and on imputed data set using three imputation methods: multiple imputation by chained equations, k-nearest neighbour imputation, and singular value decomposition imputation. We tested various criteria with different proportions of missing data (ranging from 5% to 50%) under different missingness assumptions. Q2-leave-one-out component selection methods gave more reliable results than AIC and BIC-based ones.

Formation Dominates Resorption With Increasing Mineralized Density and Time Postfracture in Cortical but Not Trabecular Bone: A Longitudinal HRpQCT Imaging Study in the Distal Radius

“…Leave-one-out crossvalidation was used to calculate the predictive power of the model due to the small number of datapoints and large number of variables evaluated. The number of model components was determined based on the Q 2 value of the cross-validation, such that additional components were only included if significant improvements to the model occurred, which were identified by increases in Q 2 of at least 0.0975 (28,29) . The variables were sorted by variable influence on projection (VIP) and the model was run iteratively including additional variables until the Q 2 score no longer improved.…”

Antecedents and Relative Importance of Student Motivation for Science and Mathematics Achievement in TIMSS

“…Participants' scores on the principal components defining the constructs were subjected to Orthogonal Projection to Latent Structures analysis (OPLS) (Trygg and Wold, 2002) for analysis of the relative importance of motivational and situational variables for predicting student achievement (Research question 1) and the relative importance of situational variables for predicting Students' motivational characteristics (Research question 2). The analyses were performed in the SIMCA P+ software (Umetrics, 2013), using the modified non-linear iterative partial least squares regression algorithm (NIPALS; Wold, 1975;Nengsih et al, 2019) for imputing missing data.…”

Denitrification in soil as a function of oxygen availability at the microscale

“…In addition to scans of the entire core, four individual aggregates (4-8 mm) of each soil were also scanned with X-ray CT (80 kv, 75 µA, 1 s exposure time, no filter, 2400 projections, two frames per projection), reconstructed in 8 bit at a voxel resolution of 5 µm, filtered with a 2D non-local means filter, and segmented into pores and background with the Otsu thresholding method (Otsu, 1975). The largest cuboid fully inscribed in an aggregate was cut and used for subsequent diffusion modelling as described below.…”