两层分类器模型应用于个人信用评估

The Two-layer Classifier Model and its Application to Personal Credit Assessment

针对不同的具体问题,传统机器学习算法的预测精度往往存在差异,而集成学习能够综合若干基分类器的预测结果,可以使得分类效果显著提升。首先,简单的介绍了集成学习的基本思想,并分析了Stacking集成算法相对于传统经典集成算法的优势;其次,基于Stacking集成框架,运用UCI的信用评估数据集,构建两层分类器学习模型用以评估个人信用;最后,将提出的模型方法用于实证分析,实验表明相对于SVM、RF、ANN、GBDT这些单一学习方法,以及对这些单一学习方法的结果进行简单的平均集成,两层分类器的Stacking集成学习的预测效果更优。

With respect to the different specific problems, the prediction accuracy of traditional machine learning methods often exist difference, while ensemble learning achieves significant improvement in classification performance by combining several of base classifiers. First, the basic idea of ensemble learning is briefly introduced, and the advantages of Stacking over the traditional classical ensemble algorithms are analyzed. Then, based on the Stacking framework, the two-layer classification model is developed to evaluate the personal credit by using the UCI datasets. Finally, the proposed method is applied to the empirical analysis, and the results show that compared with the single machine learning method of SVM, RF, ANN, GBDT and simple average ensemble, Stacking with two-layer classifier has a better prediction effect.