The performance of Taguchi’s T-method with binary bat algorithm based on great value priority binarization for prediction

Abstract

Taguchi’s T-method is a predictive modeling technique under the Mahalanobis-Taguchi system that is based on the regression principle and robust quality engineering elements to predict future state or unknown outcomes. In enhancing prediction accuracy, the T-method employed Taguchi’s orthogonal array as a variable selection approach to determine a subset of independent variables that are significant toward the dependent variable or output. This, however, leads to sub-optimality of prediction accuracy as the orthogonal array design lacks in offering higher-order variable interactions, in addition to its fixed and limited variable combinations to be assessed and evaluated. This paper proposes an optimization algorithm based on the Binary Bat algorithm methodology for replacing the conventional orthogonal array approach. Specifically, a Great Value Priority binarization scheme is employed to transform the continuous location of the bat into a binary bit, representing a combination of the variable in binary string form. A comparative study was conducted, and the mean absolute error metric was used as the performance measure. Experiments show that the T-method prediction accuracy with the Binary Bat algorithm based on the Great Value Priority binarization scheme is better than that of the conventional T-method-orthogonal array.

Keywords

• mahalanobis-taguchi system
• taguchi’s t-method
• binary bat algorithm
• great value priority
• prediction model