Towards crafting an improved functional link artificial neural network based on differential evolution and feature selection

The proposed work describes an improved functional link artificial neural network (FLANN) for classification. The improvement in terms of classification accuracy of the network is realized through differential evolution (DE) and filter based feature selection approach. Information gain theory is used to filter out irrelevant features and provide relevant features to the functional expansion unit of FLANN as an input, which in turn maps low to high dimensional feature space for constructing an improved classifier. To fine tune the weight vector of the given network, differential evolution is used. The work is validated using skewed and balanced dataset retrieved from the University of California Irvine (UCI) repository. Our systematic experimental study divulges that the performance of the differential-evolution trained FLANN is promising than genetic algorithm trained FLANN, ISO-FLANN, and PSO-BP.