Enhanced Predictive Modeling Using Multi Block Methods

Hong, JJ; Zhang, J; Morris, J

doi:10.1016/S1570-7946(09)70055-0

Enhanced Predictive Modeling Using Multi Block Methods

Lookup NU author(s): Jeong Hong, Dr Jie Zhang ORCiD, Emeritus Professor Julian Morris

Downloads

Full text for this publication is not currently held within this repository. Alternative links are provided below where available.

Abstract

Multi block PLS (MBPLS) models have been reported to handle the complexity of data which is difficult to anlayse with the conventional PLS models. However, the conventional MBPLS models do not offer improved predictive modeling in terms of prediction performance. A data partitioning method for enhanced predictive process modeling is proposed in this paper and is called Time-MBPLS. It enables data to be separated into blocks by different measuring time. Model parameters can be used to express contributions for all variables in a certain time period and it is possible to determine data blocking structures according to variable impacts on the quality variables. The proposed method is applied to the inferential estimation of a quality variable in a bio-production process. The results demonstrate that the proposed method can improve model prediction performance on the unseen testing data. © 2009 Elsevier B.V. All rights reserved.