Research On Extraction And Prediction Of Data Feature Based On Deep Learning

It is because that here comes the big data era,big data become a valuable resources.There is a lot of useful information in the data,such as the law of disease transmission,human behavior,marketing and so on.Data prediction,as one of the core applications of data mining,is an important means to develop and utilize the data resources.Data features are often used to describe data properties and relationships,as an English saying” The better the data feature,the better the prediction”.Therefore,it is a very important and meaningful work to extract the better data features for the data with the characteristics of nonlinearity,instability and uncertainty.But,most of the existing feature extraction algorithms and prediction models are based on the shallow model now.The data features obtained from these models are calculated on the basis of statistical methods,so that the internal relationship between sample data could not be expressed accurately and effectively by these algorithms.The development of deep learning(DL)could not be separated from the bionics of the brain visual system.The data samples can be abstracted layer by layer with a deep-level architecture model in the DL,and we could get deep and better data features.DL has been widely used in the field of image recognition,speech processing,natural language representation and any others for its strong advantage in the extraction and transformation of data features.Therefore,this paper focuses on the data feature extraction and prediction on the study with deep learning,and the specific work which has done could be as follows:(1)A time series data prediction problem.Based on the preprocessing of hospital outpatient data,a prediction model which is based on the recurrent neural network and restricted Boltzmann machine(RNN-RBM)deep structure is constructed.It combines the advantages of RNN and RBM neural network to the timing data processing,and uses the deep structure to study the features of the outpatient data layer by layer.The comparison experiment proves the effectiveness of the forecasting model.(2)A community monitoring algorithm in a complex network.In order to solve the problem that the old community detection algorithm which is based on the network module degree could not extract non-linear network features,a feature learning algorithm based on deep learning is proposed.The algorithm can express network topology by the non-linear features.The result on synthetic datasets and real world datasets show that the feature learning method based on deep learning is real effective.(3)The prediction of community evolution of complex networks.The dynamic community evolution contains the information of the community topology and the evolutionary time series of the community.So this paper want to choose some better features of community evolution from the inside and outside of communities,and the feature is entered into the neural network based on deep learning,the neural network is used for the transformation and learning of features.The prediction model can be used to obtain the representative features of the data,which can improve the accuracy of prediction results.