Research On Imbalanced Data Classification Based On Neural Network Adversarial And Ensembles

Data classification is an important means to obtain information and value from data by machine learning and data mining.Traditional data classification algorithms are usually applied to data sets with balanced data probability distribution.However,in real life and industrial production,many data sets are imbalanced,small number of types of data usually have more important information,while data is misclassified,a high price would be payed,such as medical diagnosis,credit card fraud detection.In this case,the traditional classification algorithms with the overall classification accuracy rate are not suitable for solving this type of problem.In order to solve the problem of imbalanced data classification,A deep neural network(DNN)based on evolutionary algorithm,a deep generative adversarial network(DGAN)based on noise auto-encoder and a deep neural network ensemble are proposed as the advantage of DNN to extract the potential features in complex problems.The main results and innovations of this paper are as follows:An imbalanced data classification model based on DNN is proposed.The significant features difference in imbalanced data is extracted by deep auto-encoder(DAE)and deep Boltzmann machine(DBM).In this model,the water wave optimization algorithm(WWO)is applied to optimize the structure and parameters of the DNN,and improves the performance of the DNN,and then increases the classification accuracy.An adversarial deep denoising autoencoder(GAN-DEA)is proposed.Through the confrontation training of generator and discriminator,the features of positive and negative samples in imbalanced data are obtained,and the trained generator is used to generate minority samples to balance the data.The proposed model is applied to bank fraud transfer detection,which has achieved a good classification effect and produced a larger economic benefits..The deep neural network ensemble based on meta-heuristic algorithm optimization solves the problem of imbalanced data classification in practical problems.In this model,we apply the GAN-DEA as a member neural network,and use genetic algorithm to adjust the weight between each member neural network.Experiments prove that proposed neural network ensemble has a good performance in solving the problem of imbalanced data classification.