Feature selection with particle swarms

摘要: Feature selection is widely used to reduce dimension and remove irrelevant features. In this paper, particle swarm optimization employed select feature subset for classification task train RBF neural network simultaneously. One advantage that both the number of features configuration are encoded into particles, in each iteration PSO there no iterative training sub-algorithm. Another fitness function considers three factors: mean squared error between outputs desired outputs, complexity features, which guarantees strong generalization ability network. Furthermore, our approach could as small-sized possible satisfy high accuracy requirement with rational time. Experimental results on four datasets show method attractive.