Improving fuzzy neural networks using input parameter training

Fuzzy neural networks allow the implementation of rules in a neural topology and therefore make it possible to add knowledge to neural systems. An overview of applying fuzzy neural networks to financial problems has been given by the author (Proc. NAFIPS ´97). In this paper an additional improvement is given, which speeds up training in forecasting, and which can improve network performance. Normally the inputs to a neural network are technical indicators; this is better than showing raw data to the network. The optimisation of the parameters necessary for these indicators is a separate operation from the weight training and topology optimisation. In the approach presented the optimisation of these parameters is included into the weight training stage, thus removing one level of optimisation