3.1. Process parameters and process response identification The idea of using NNs to model the processes is to create networks that take process parameters as inputs and produce process responses (such as quality characteristics) as outputs. One way to do this is to assign all available process parameters as network inputs, and then let the network adjust itself during training so that the connection of any insignificant parameters becomes weak. Another approach is to be more selective, and assign as inputs only those parameters that are believed to influence the process outputs. The first approach has been termed the ‘‘global network’’ while the second is called the ‘‘focused network’’ (Wilcox and Wright, 1998). These authors showed that, when modelling the same process, the focused network performed better than the global one, suggesting that process parameters should be carefully selected to improve the performance of a network. However it should be noted that, for either approach if a significant process parameter input is missing, the network performance will be compromised, as the variation in the available input data will not be sufficient to explain the variation in the process output characteristics. Process response used for NN training should be an appropriate measurement of the success of the interested process. The selection of the suitable response requires understanding of the process. NN can be trained to predict both single-response and multi-response.