基于最优基模型集成算法的信贷违约预测研究

Study on credit default prediction based on optimal base model ensemble algorithm

为了保障金融机构的金融安全,应用机器学习进行信贷违约预测已成为研究重点。为此,构建了6个机器学习基模型,调至最优参数后再分别用Voting、Stacking、Adaboost方法集成。实验表明,在多种基模型中,随机森林(RF)取得了较好的效果;而在集成方法中,Adaboost对基模型的提升最显著。文中构建的Adaboost-RF模型在信贷预测上的交叉验证得分达到了0.904,明显优于其它方法,该方法对金融机构制定信贷决策具有一定的借鉴意义。

In order to ensure the financial safety of financial institutions,the application of machine learning in credit default prediction has become a research focus.To this end,six machine learning base models are constructed,and after tuning to optimal parameters,they are integrated separately using Voting,Stacking and Adaboost methods.The experiment shows that among multiple base models,the Random Forest(RF)achieves better results;while in the ensemble methods,Adaboost had the most significant improvement on the base models.The Adaboost-RF model achieves a cross-validation score of 0.904 in credit prediction,which is significantly better than other methods,and this method has certain reference value for financial institutions in making credit decisions.