Imbalanced Classification Of Structured Data Based On Generative Adversarial Nets

There are a large number of imbalance data,such as credit card fraud data,rare disease data,and so on.One main challenge for machine learning research is to learn a classifier from imbalanced dataset,since they will be biased towards the features of the majority group,which makes it hard to identify the minority group.The existing methods for solving the imbalance classification problem are mainly divided into two aspects: data level and algorithm level.The data level method makes the data set relatively balanced by changing the number of samples in different categories.The algorithm level mainly includes cost-sensitive learning,boosting,etc.By increasing the weight of the small sample,the classifier is more in the training process.Focus on a small number of samples.This paper proposes an imbalanced classification method based on Generative Adversarial Nets for Structured data.In this paper,the WGNA-GP is trained through the minority class samples,then we put the random noise into the trained WGANGP model to output the synthetic sample,this synthetic samples is added to the imbalanced data set to obtain the multi-group balanced data set.Based on the multigroup balanced data set,we can obtain and optimized multiple classifiers,from this multiple classifiers,we choose the best classifier as the final classifier.In this paper,the real online payment behavior data is used to verify the proposed method,and choose Precision,recall(TPR),TNR,F-measure and G-mean are as the indicator to measure the performance of the imbalanced classification.At the same time,the SMOTE algorithm and the Bootstrap algorithm are used as comparisons experimental.The experimental results show that,the proposed method has a better performance.