Missing Data Imputation In Multi-task Generative Adversarial Imputation Nets

Nowadays,the rapid development of big data technology has swept the whole field of scientific research,and data,the basis of which can be used flexibly,plays a crucial role.Complete data accumulation can make scientific research even more powerful.However,although the existing data sets are numerous,it is inevitable that there are some deficiencies in the data,especially for some data with high cost and long cycle,which is a common phenomenon,so there are few data sets that can be directly used for scientific research.In view of the dilemma of incomplete data sets,relevant scholars have carried out a series of data complement methods to solve most of the related problems.Among them,the method of using neural network to supplement data is easy to understand,fast and accurate,which provides another important solution for data imputation.In this paper,we use the current popular generative adversarial network to fill the missing value of data,and we will Yoon’s unsupervised method(GAIN)published in 2018 extends to supervised learning method(MT-GAIN).On the one hand,on the basis of the original generator(G)and discriminator1(D1),a prompt mechanism(H)is added to the network structure,which can ‘induce’the generator to plant in the direction of real data,and the discriminator2(D2)combines with data label to raise The information provided corrects and supervises the data after imputation,and guides the data to replant in the direction that conforms to the data law and distribution law.In this way,they supervise and promote each other and cooperate to replant the missing data.On the other hand,this method makes use of supervised learning and multi task thinking,makes full use of data information,and can be used for replanting in the case of missing data of one or more attributes.In addition,a comparative experiment is carried out on the existing data missing methods(GB-FS-EL(integrating imputation,genetic-based feature selection and ensemble learning),KNN,MICE,EM,GAIN).As for the accuracy of replanting and classification accuracy,the method in this paper is more traditional According to the advantages of replantation method,especially when the data loss is high,it has a good effect.