Wrapped feature selection by means of guided neural network optimisation
Bart Baesens, Stijn Viaene, Jan Vanthienen, Guido Dedene · 2002
We discuss the implementation of a wrapped neural network feature selection approach, introduced here as the weight cascaded retraining (WCR) algorithm. The paper provides an outline of the algorithm and elaborates on its formal underpinnings. Central to the whole feature pruning approach is the iteratively conceived guided function optimisation realised by passing the optimised weight vector from one iteration step to the next. This essentially gives rise to a cascaded form of neural network retraining. The theoretical exposition of the WCR algorithm is illuminated and benchmarked by means of the publicly available UCI case material. It is illustrated that WCR based neural network feature selection may be very effective in reducing model complexity for classification modelling via neural networks.