Variable group selection based on regression trees: Paper machine case study

This paper presents a methodology for selecting best groups of predictor variables based on regression trees. Test results of the developed methodology applied to industrial pilot paper machine data are presented. Specifically, the results list process variable groups, which are more valuable in predicting paper quality variables. The benefit of paper quality prediction based on process variables is the timely reaction to changes happening during production process and, thus, the reduced operational costs. The proposed regression trees based group variable ranking methodology shows stable results on both data sets used in this study.