ENBIS-16 in Sheffield

11 – 15 September 2016; Sheffield Abstract submission: 20 March – 4 July 2016

Factor Screening in Non-Regular Designs with Restriction on Parameters

13 September 2016, 10:10 – 10:30

Abstract

Submitted by
Shahrukh Hussain
Authors
Shahrukh Hussain (Department of Mathematics, Norwegian University of Science and Technology), John Tyssedal (Department of Mathematics, Norwegian University of Science and Technology)
Abstract
Factor screening plays an important role in industrial experimentation where we usually have large number of factors with their interactions and small number of runs. The goal is to find best subset of predictors that have a great influence on response of interest. In this paper we proposed an efficient variable selection algorithm to screen active effects in non-regular designs by using their projection properties. The proposed method is inspired by (Tyssedal & Hussain, 2016) to screen active factors. On each projection model forward regression strategy with change in coefficient of determination as test statistic proposed by (Tyssedal & Hussain, 2016) to screen active factors is conducted by restricting the number of terms to enter the model. Then gradually reducing the number of possible subsets of active factors. Empirical comparison with traditional forward selection approaches used in projection model is done. Our simulation study shows that proposed method compares favorably well with traditional forward selection approaches when some restrictions are imposed.

Return to programme