Feature Selection and Optimization of Back Propagation Neural Network Parameters Using the Differential Evolution Algorithm

Back propagation neural network is successfully used in various fields, particularly in pattern recognition. Despite numerous applications, back propagation neural network`s design and optimization are developed by trial-and-error process, which is time-consuming. On the other hand, although a dataset may contain many features, these features may not be useful in a back propagation neural network. Therefore, differential evolution based algorithm, denoted as DE+BP, and is proposed to determine the number of neurons in the hidden layer, learning rate, momentum rate, and the feature selection for the back propagation neural network. The DE+BP combine benefits of global search by differential evolution algorithm and local search by back propagation algorithm. The results of the proposed algorithm on 7 data sets of the UCI machine learning repository have been compared with 22 classification algorithms and related work in other papers. The experimental results showed high classification accuracy of the proposed algorithm compared to other algorithms.