Feature Selection Combined Classificatory Concentration with Improved RBF Neural Network

Feature selection is the core research topic in text categorization. Firstly, an optimized document frequency and classificatory concentration were presented. And then, an adaptive quantum-behaved particle swarm optimization (AQPSO) algorithm was provided to improve the performance of radial basis function (RBF) neural network. By applying AQPSO algorithm to train the central position and width of the basis function adopted in the RBF neural network, and computing the weights of the network with least-square method (LSM), the RBF neural network was improved. Finally, a combined feature selection method was proposed. The combined feature selection method firstly uses classificatory concentration to filter out some terms to reduce the sparsity of feature spaces, and then employs the improved RBF neural network to select more outstanding feature subset. The experimental results show that the combined method is excellent in accuracy rate and recall rate.