Abstract
Taguchi parameter design is a quality approach to design better products and processes, less sensitive to changes of the environmental and productive conditions. Robustness against changes in factors affecting processes is the key concept. Some recent papers have used a two steps methodology to improve parameter design. The first step determines the objective function using Artificial Neural Networks (ANN) to predict the value of the response variable when factors are in some specific levels (different to those included in the experiments). The second step looks for the optimal parameter combination. Our proposal here is centered in improving the first of these two steps, and consists in the development of new systems to model the response variable, based in Classification and Regression Trees (CART) and in Random Forest (RF), as an alternative to ANN and with the aim of creating a more robust strategy.
