Group Lasso based redundancy-controlled feature selection for fuzzy neural network

If there are a lot of inputs, the readability of the "If-then" fuzzy rule is reduced, and the complexity of the fuzzy neural network structure will be increased. Hence, to optimize the structure of the fuzzy rule based neural network, a group Lasso based redundancy-controlled feature selection (input pruning) method is proposed. For realizing feature selection, the linear/nonlinear redundancy between features is considered, and the Pearson's correlation coefficient is employed to construct the additive redundancy-controlled regularizer in the error function. In addition, considering the past gradient information, a novel parameter optimization method is presented. Finally, we demonstrate the effectiveness of our method on two benchmark classification datasets.