联合迁移学习和强化学习的不平衡分类方法

Unbalanced classification method combining transfer learning and reinforcement learning

在不平衡二分类问题中,为避免上采样算法的生成样本质量参差不齐导致模型分类性能较差的问题,提出一种联合迁移学习和强化学习的不平衡样本分类模型。基于上采样方法对少数类进行样本生成,将生成样本集看作源域,原始不平衡数据集看作目标域,计算源域中每个样本在不平衡分类问题上的贡献度,得到源域的先验知识;建立强化学习智能体,智能体利用先验知识对生成样本进行筛选,将贡献度大的样本挑选进训练集;利用新建立的训练集训练分类器,实现不平衡样本分类。实验结果表明,在7个不平衡数据集上,相较于现有算法,所提模型均能在一定程度上提高分类效果。

In the unbalanced binary classification,to avoid the problem of the poor classification performance of the model due to the uneven quality of the generated samples created using the up-sampling algorithms,an unbalanced samples classification model combining transfer learning and reinforcement learning was proposed.Samples for minority class were generated based on the upsampling method.The generated samples set was regarded as the source domain,and the original untalanced data set was regarded as the target domain.The contribution degree of each sample in the source domain on the unbalanced classification problem was calculated to obtain the prior knowledge of the source domain.A reinforcement learning agent was established.The agent selected the samples with large contribution into the training set with prior knowledge.The newly established training set was used to train the classifier and the unbalanced sample classification was realized Compared with the existing algorithms,experimental results on seven unbalanced data sets show that the proposed model can improve the effects of classification to some extent.