Screening important design variable for building a usable model: genetic algorithm-based partial least-squares approach

Abstract

This study proposes a method of screening product design variables before building usability models. The proposed method finds a set of product design variables to minimize the root-mean-squared error (RMSE) of partial least-squares regression (PLSR) models that are used as alternatives when the number of variables is too large to build multiple linear regression models. A genetic algorithm is applied to the minimization process (called GA-based PLS). Selected variables are used to build usability models based on a multiple linear regression technique. Other variable screening methods such as expert opinions, principal component analysis (PCA), cluster analysis, and partial least squares (PLS) are also applied to compare the performance of the proposed method. The results show that the usability models using the variables screened by the GA-based PLS are one of the best models in terms of prediction capability, model stability, and the number of variables.