基于集成学习的中国P2P网络借贷信用风险预警模型的对比研究*

操玮, 李灿, 贺婷婷, 朱卫东

Predicting Credit Risks of P2P Loans in China Based on Ensemble Learning Methods

Cao Wei,Li Can,He Tingting,Zhu Weidong

表4 根据三个指标的集成模型准确性评估 (训练测试比为60:40)

集成算法 基分类器 准确率(%) Type-I error (%) AUC A FS A FS A FS Bagging LR 97.17 98.79 7.47 6.71 0.987 0.989 CART 98.08 98.58 6.32 2.87 0.936 0.999 C4.5 97.67 98.28 6.90 2.87 0.949 0.998 MLP 96.06 98.08 10.92 8.05 0.988 0.998 SVM 97.47 97.97 9.20 6.32 0.973 0.995 Boosting LR 97.98 98.48 5.74 3.45 0.984 0.898 CART 97.17 98.88 6.32 5.17 0.983 0.994 C4.5 97.17 98.07 5.17 4.02 0.995 0.996 MLP 97.27 98.88 8.05 2.30 0.995 0.999 SVM 97.97 98.28 8.62 4.59 0.954 0.971 Random Subspace LR 95.55 97.27 15.52 8.05 0.984 0.996 CART 96.46 97.17 4.59 2.87 0.940 0.981 C4.5 95.05 98.07 6.32 4.02 0.961 0.994 MLP 95.45 97.67 8.62 7.47 0.965 0.997 SVM 96.87 97.37 9.77 7.47 0.955 0.963 Rotation Forest LR 98.68 99.69 3.45 0.57 0.998 1.000 CART 98.48 99.19 3.45 1.15 0.992 0.998 C4.5 97.97 98.99 6.90 5.17 0.954 0.996 MLP 98.07 99.29 6.32 1.15 0.997 1.000 SVM 98.78 99.79 4.59 0.00 0.975 0.999