基于BLT方法的样本不平衡分类研究

Research on Sample Imbalance Classification Based on BLT Method

针对样本数量分布不平衡的分类问题,使用分支学习树(BLT)的方法来提升分类精度,并应用于肿瘤免疫亚型分类问题,从而验证其有效性。统计每种免疫亚型的数量,以此建立一棵哈夫曼树,用传统分类器作为分支节点,进行自顶向下逐步分类方式,实现对不平衡数据的准确分类。使用BLT方法后,对比传统分类器分类准确率提升1.5%左右,在误分最严重的类别上,分类性能提升最高可达79%。上述方法可用于提升样本不平衡的分类问题的分类性能,且在样本数量较少的类别上效果尤为明显。

Aiming at the classification problem with unbalanced sample size distribution,the Branch Learning Tree(BLT)method is used to improve the classification accuracy and applied to the classification of tumor immune subtypes to verify its effectiveness.Count the number of each immune subtype,and build a Huffman tree based on it,use traditional classifiers as branch nodes,and perform a top-down gradual classifi⁃cation method to achieve accurate classification of unbalanced data.Results:After using the BLT method,the classification accuracy of the original classifier was improved by about 1.5%,and the classification performance was improved by up to 79%in the most misclassified cat⁃egory.The above method can be used to improve the classification performance of the classification problem with unbalanced samples,and the effect is particularly obvious on the categories with a small number of samples.