一种基于密度的SMOTE方法研究

重采样技术在解决非平衡类分类问题上得到了广泛的应用。其中，Chawla提出的SMOTE（Synthetic Minority Oversampling Technique）算法在一定程度上缓解了数据的不平衡程度，但这种方法对少数类数据不加区分地进行过抽样，容易造成过拟合。针对此问题，本文提出了一种新的过采样方法:DS-SMOTE方法。DS-SMOTE算法基于样本的密度来识别稀疏样本，并将其作为采样过程中的种子样本；然后在采样过程中采用SMOTE算法的思想，在种子样本与其k近邻之间产生合成样本。实验结果显示，DS-SMOTE算法与其他同类方法相比，准确率以及G值有较大的提高，说明DS-SMOTE算法在处理非平衡数据分类问题上具有一定优势。

Research on the SMOTE method based on density

In recent years, over-sampling has been widely used in the field of classification of imbalanced classes. The SMOTE(Synthetic Minority Oversampling Technique) algorithm, presented by Chawla, alleviates the degree of data imbalance to a certain extent, but can lead to over-fitting. To solve this problem, this paper presents a new sampling method, DS-SMOTE, which identifies sparse samples based on their density and uses them as seed samples in the process of sampling. The SMOTE algorithm is then adopted, and a synthetic sample is generated between the seed sample and its k neighbor. The proposed algorithm showed great improvement in precision and G-mean compared with similar algorithms, and it has advantage of treating imbalanced data classification.