【摘要】 针对信用卡交易数据的不平衡重叠问题，提出一种基于生成对抗网络的端到端一类分类方法。提出一种基于PCA和T＿SNE的混合数据降维方法，对清洗后的数据进行特征降维；将降维后的数据送入所提出的基于LSTM和aMLP的生成对抗网络(aLMGAN),提出一种基于闵可夫斯基距离(Minkowski distance)的损失函数(Min-loss)代替原始生成对抗网络中的交叉熵损失函数，对正常交易数据进行单类稳定训练，形成一种特殊特征模式，区分不属于该特征的异常数据。通过使用kaggle上两个真实的公共信用卡交易数据集进行实验，验证了aLMGAN算法的有效性。更多还原

【Abstract】 Aiming at the imbalanced overlapping problem of credit card transaction data, an end-to-end one class classification method based on generation countermeasure network was proposed. A method based on PCA and T＿SNE hybrid data dimension reduction method was proposed, which reduced the dimension of the cleaned data. The reduced dimension data was sent into the proposed LSTM and aMLP based generation countermeasure network(aLMGAN), and a Minkowski distance based loss function(Min-loss) was proposed to replace the cross entropy loss function in the original generation countermeasure network, and single class stability training was conducted for normal transaction data to form a special feature mode to distinguish abnormal data that did not belong to this feature. By using two real public credit card transaction datasets on kaggle, the experiment verifies the effectiveness of aLMGAN algorithm.更多还原