GAN-based semi-supervised for imbalanced data classification

Most of the traditional classification algorithms are based on the premise that the datasets are uniformly distributed or roughly equivalent. Once the sample dataset is not balanced , the classification performance drops sharply. To efficiently deal with the imbalance of data, an improved generative adversarial network (GAN) algorithm is proposed in this work . Firstly, we construct artificial samples so that more minority-class's data can be obtained via optimizing GAN loss function. Secondly, we build a fully-connected network for structured data classification. Finally, experimental evaluations are conducted on two open structured-datasets and the results of the proposed algorithm demonstrate a good applicability for the classification of structured data.