BP Neural Network Feature Selection Method Based on Sensitivity Analysis

The data for the analysis of data may contain hundreds of features. Many of the features are irrelevant in data mining, So it is particularly important to find out the minimum set of features to improve the efficiency of data mining. A method of neural network features selection based on sensitivity analysis is presented in the paper. It avoids the deficiency of traditional neural network methods that needs to train a network using all features. It ranks the features of initial features set by using the method of sensitivity analysis, and then removes the secondary features to achieve dimension reduction. The feathers are selected by the BP neural network at last. The simulation results show the efficiency of this approach.