不平衡类分类问题的逻辑判别式算法

Logistic Discrimination Algorithms for Imbalance Classification Problems

针对不平衡分类问题,提出了逻辑判别式算法.该算法使用拟牛顿法迭代求解模型参数,考虑模型的准确率和召回率,构造了新损失函数(Likelihood Estimation and Recall Metric,LERM);设计了用于不平衡类问题的逻辑判别式算法(Logistic Discrimination Algorithms for Imbalance,LDAI).16个数据集上的实验结果表明,与传统的逻辑判别式、基于过采样和欠采样的逻辑判别式相比,LDAI模型在召回率、f-measure、g-mean等指标上都表现出明显优势.

Logistic discrimination algorithms was applied to class-imbalance problem. In this algorithm,the iterative quasi-newton method was used to solve the model parameters. Taking both the model accuracy and recall-rate into consideration,the LERM（ Likelihood Estimation and Recall Metric） was constructed and the logistic discrimination algorithms for imbalance（ LDAI） was designed to figure out the imbalance problems.Experimental results on 16 data sets showed that the LDAI performed significantly better than traditional logistic discrimination,under-sampled and oversampled logistic discrimination on the properties of recall-rate,f-measure,g-mean and so on.