基于DESMID-AD动态选择的类别不平衡信用评估模型

针对现实信用评估业务中样本类别不平衡和代价敏感的情况,为降低信用风险评估的误分类损失,提出一种基于DESMID-AD动态选择的信用评估集成模型,根据每一个测试样本的特点动态地选择合适的基分类器对其进行信用预测.为提高模型对信用差客户(小类)的识别能力,在基分类器训练前使用过采样的方法对训练数据作类别平衡,采用元学习的方式基于多个指标进行基分类器的性能评估并在此阶段设计权重机制增强小类的影响.在三个公开信用评估数据集上,以AUC、一型、二型错误率以及误分类代价作为评价指标,与九种信用评估常用模型做比较,证明了该方法在信用评估领域的有效性和可行性.

Class-imbalance credit scoring using DESMID-AD

In view of class-imbalanced and cost-sensitive problem in real credit scoring business, as well as financial institutions hope to decrease cost-loss value, this paper proposed dynamic ensemble selection using meta-learning for imbalanced data according to accuracy and diversity(DESMID-AD) for class-imbalance credit score. In order to improve the prediction accuracy of minority sample, before generation of the classifiers pool, it used an over-sampling method to balance the training set. Next, it employed the meta-learning to evaluate whether the base classifier can correctly classify a test sample and proposed a weighting mechanism to weight the instances in a competence region at this stage. Experimental comparisons of three imbalanced credit score datasets suggest that this model exhibits the best overall credit-scoring performance compared with nine well-known credit scoring models and reveal the effectiveness and feasibility of the proposed method in the field of credit scoring.