The Study On The Adaptive Elastic Net Neural Network Model And Algorithm

With the development of artificial intelligence,artificial neural network has become a research hot spot.Because it has excellent fitting ability and can theoretically fit any data distribution,artificial neural network is widely used in many fields.In addition,it has achieved many good results in image classification,voice recognition and natural language processing,etc.However,neural network model will also appear overfitting due to the high complexity of the model.In other words,neural network model works well on the training sets,but the generalization of the neural network model on test datasets is not good enough.To solve overfitting problem,many methods have been proposed.Regularization methods is one of the methods with wide range of applications.Some regularization models have been proposed,and achieved favorable results.However,these models can not satisfy the full properties of the ideal regularization model proposed by Antoniadis.Base on this,this paper proposes adaptive elastic net neural network model.The model combines a new regularization term with neural network model.This paper proves that adaptive elastic neural network model has good properties in theory.Besides,we have proved that this model can solve overtitting problem better through a series 'of experiments.Moreover,in imbalanced problem,we find that the neural models tend to overfit the majority class samples.Then we have done a series of experiments,to verify that the adaptive elastic net neural network model can also get good results on three different indicators in solving class imbalance problem.The contributions of this paper are as follows:First,we combine adaptive elastic net methods with neural network model.An adaptive elastic net neural network model and adaptive momentum optimization algorithm based on adaptive elastic net are proposed.Next,adaptive elastic net neural network model has oracle property and group effect is proved in theory.And the process of proving is given.Then,we perform experiments on classification and regression datasets,and verify the ability of adaptive elastic net neural network model to solve overfitting is better than L1,L2 and elastic net regularization model.Finally,adaptive elastic net neural network model is used to solve class imbalance problem.We perform a series of experiments on multi-class and binary classification datasets and verify this model can solve class imbalance problem effectively.In addition,adaptive elastic net neural network model can achieve better result than random oversampling,random undersampling and synthetic minority oversampling technique(SMOTE).