A STRUCTURE-ADAPTIVE APPROACH FOR NEURAL-NETWORK-BASED FEATURE SELECTION

Feature selection is an important part of data processing. Conventional approaches ignore the change of number of hidden units, so network architecture often becomes irrational in the process of feature selection. It hinders further selection of features and further improvement of network generation ability. To solve the above problem, a structure adaptive approach for neural network based feature selection is proposed in this paper. By pruning the redundant input features and hidden units alternatively, network architecture is kept reasonable. Experiments show that this method can effectively select features while improving the generalization ability of network.