At present,the methods of feature selection based on random forest algorithm mostly aim at optimizing the overall classification accuracy.However,unequal misclassification cost of imbalance data is widespread in the credit risk assessment process.At this moment,it is unsuitable to use the precision to make the classification performance evaluation index.The AUC value of area under the ROC curve was used as the classification performance index of the binary classification algorithm to construct a feature selection algorithm AUCRF based on the random forest algorithm.The empirical analysis of the Australian credit data in the UCI machine learning database was carried out.The results showed that the model based on AUCRF algorithm obtained higher classification performance with smaller feature subset,AUC =0.934 6.Therefore,the AUCRF algorithm can be used in the credit risk feature selection with the unequal misclassification cost.%目前基于随机森林算法的特征选择方法多以优化总体分类精度为目标.然而,信用风险评价过程中错分代价不对等的不平衡数据广泛存在.此时,用精度作分类性能评价指标不合适.采用ROC曲线下面积AUC值作二分类算法的分类性能指标,构造一个基于随机森林算法的特征选择算法AUCRF,并对UCI机器学习库中的澳大利亚信用数据进行实证分析.结果表明,基于AUCRF算法的模型能以较小的特征子集获得较高的分类性能,AUC=0.934 6.因此,AUCRF算法可用于错分代价不对等的信用风险特征选择.
展开▼
