改进的SSVM集成算法在信用风险评估中的应用

Application of improved SSVM integration algorithm in credit risk assessment

为进一步提高选择性支持向量机(SSVM)的分类精度,提出一种基于改进的SSVM的集成算法(AR-SKB)。利用AdaBoost算法和基于广义差别矩阵的粗糙集属性约简算法对样本和样本的属性特征进行扰动,生成差异度较大的个体SVM;利用自组织映射(SOM)和K-means聚类算法结合的聚类算法(SOM-K)对训练出来的个体SVM进行分类,选择每类中训练精度最高的SVM作为最优个体;用BP算法将最优个体进行非线性集成。实验结果表明,该算法在UCI两个数据集上的分类精度分别提高了2.7%和2.2%。

To further improve the classification accuracy of selective support vector machine(SSVM),an improved algorithm based on SSVM(AR-SKB)was proposed.The AdaBoost algorithm and the rough set attribute reduction algorithm based on generalized difference matrix were used to perturb the attribute features of samples and samples to generate individual SVM with large difference.The self-organizing map(SOM)and K-means clustering algorithm(SOM-K)were used to classify the trained individual SVM,and the SVM with the highest training accuracy was selected as the optimal individual.The optimal individual was integrated using BP algorithm.Experimental results show that the classification accuracy of the proposed algorithm on the two UCI data sets is increased by 2.7% and 2.2% respectively.