Use of partial least squares regression for variable selection and quality prediction

Abstract

Process engineers are often eager to find the optimal levels of process variables that make the key quality variable as close to its target as possible. The quality of products is affected by a few hundreds to thousands of variables. So, it is difficult to construct a reliable prediction model from the data of many variables and small observations. The selection of important variables becomes a crucial issue naturally as well. In this paper, we introduce the partial least squares (PLS) regression for quality prediction and its use for the variable selection based on the variable importance. Some simulation results for the proposed variable selection method are presented. Further, we introduce the interval selection method based on the PLS. The variable selection procedure under PLS are then applied to several real cases.