一种基于分类算法集成学习模型的金融信贷违约预测

An Ensemble Learning Model for Predicting Financial Credit Default Based on Classification Algorithm

摘要: 随着金融市场的不断发展，金融信贷业务的激增也导致了信用风险的不断增加。为了应对这一挑战，传统的风险评估方法已经不能满足实际需求。目前集成模型成为违约问题研究的热点，通过整合多个分类算法的预测结果，充分利用各个算法的优势，以提高预测准确性和鲁棒性。本文研究了双阶段异构堆叠集成模型(DH-SEM)在金融信贷违约中的应用。该模型包括两个关键阶段，在第一阶段，选择了SVM、KNN、朴素贝叶斯作为三个监督基础学习器；在第二阶段，采用了随机森林作为元学习器来预测分类结果。对于金融信贷违约预测，DH-SEM模型预测准确率为0.886，相比传统的模型预测的更加准确。

Abstract: With the continuous development of the financial market, the proliferation of financial credit business has also led to a continuous increase in credit risk. In order to cope with this challenge, traditional risk assessment methods can no longer meet the actual demand. Currently integrated models have become a hotspot in the study of default problems, which make full use of the advantages of each algorithm by integrating the prediction results of multiple classification algorithms in order to improve the prediction accuracy and robustness. In this paper, we study the application of two-stage heterogeneous stacked integration model (DH-SEM) in financial credit default. The model consists of two key phases; in the first phase, SVM, KNN, and plain Bayes are selected as the three supervised base learners; in the second phase, Random Forest is employed as the meta-learner to predict the classification results. For financial credit default prediction, the prediction accuracy of DH-SEM model is 0.886, which is more accurate compared to the traditional model prediction.