【摘要】 准确评价P2P网贷借款人信用水平是P2P网贷平台降低借款人违约率、控制整体信用风险的基石。针对网贷借款人数据量大、维度高的特点，提出一种核非负矩阵分解与贝叶斯优化结合的Xgboost分类算法。首先利用核非负矩阵分解方法对借款人数据降维，然后将贝叶斯思想引入Xgboost方法，寻找使分类精度最高的参数组合以优化分类器性能，提高借款人信用评价准确率。仿真实验表明，该种改进的Xgboost算法，相较于经验值定参Xgboost算法及传统支持向量机算法，具有更高的分类精度。更多还原

【Abstract】 Accurate evaluation of creditworthiness of P2P online borrowers is the cornerstone of credit risk management of P2P online platforms.Based on large volume and high dimensions of loan borrowers′data,this paper proposes an Xgboost classification algorithm combined with kernel non-negative matrix factorization and Bayesian optimization.Firstly,borrowers′data dimension is reduced using kernel non-negative matrix factorization.Then the Bayesian method is introduced to optimize parameters of the Xgboost classifier to improve the accuracy of borrowers′credit assessment.Simulations show that the improved Xgboost algorithm has better classification accuracy than either the Xgboost algorithm with empirical parameters or the support vector machine algorithm.更多还原