Determination of optimal unit space data for taguchi’s t-method based on homogeneity of output

Abstract

Taguchi’s T-method is a prediction model introduced by Genichi Taguchi under the Mahalanobis- Taguchi System to determine the future state or unknown output based on existing or historical data. The prediction model was constructed using normalized signal data involving subtraction of average value of unit space data from signal data. The objective of this research is to determine a group of data having homogeneous characteristics from a densely populated region in a dataset to functioned as a basis for unit space data selection in T-method for predicting an accurate outcome. Histogram was utilized as a tool in representing data in multiple groups and a group with highest data frequency defined as unit space data. Nine different number of bins was used in assessing the effect of unit space data towards prediction accuracy. The result from the experiments on six different datasets indicates that no single number of bin fit for all in offering an optimal result. In addition, the size of unit space data and signal data do not significantly affect the final outcome. However, except for Auto MPG dataset, all different numbers of bin resulted in better prediction accuracy with less MSE and RMSE as compared to conventional T-method.