Abstract: To improve the risk control ability of P2P loan platform,a risk evaluation approach based on an improved decision tree algorithm with multi-features fusion and cost sensitive characteristic was proposed.The credit feature of borrowers was constructed from the information provided by the platform,and the feature set which was most relevant to the risk evaluation and had the least redundancy was filtered out by constructing a feature selection algorithm with AUC maximization and mutual information minimization as the objectives and integrating personal credit feature,historical borrowing records with project features.The risk evaluation of loan projects based on cost-sensitive decision tree algorithm was realized.Experiment showed that the proposed approach had better performs compared with the related work and some traditional approach.

Key words: P2P loan platform, decision tree algorithm, cost sensitive, imbalanced dataset

摘要： 为了提高P2P网络借贷平台的风险控制能力,提出一种基于多特征融合的、代价敏感决策树的网络借贷项目风险评估方法。该方法首先从互联网借贷平台上提供的借款人信息中构建借款人信用特征,然后综合借款人的历史借款信息、个人信用特征及项目特征,通过构建ROC曲线下面积最大化、互信息最小化为目标的特征选择算法,选择出与项目风险评估最相关且冗余性最少的特征集合,再通过改进的基于代价敏感的决策树算法对借款项目进行风险评估。实验表明,该方法与相关文献中提到的方法以及一些传统分类器相比,在P2P网络借贷风险评估上有更优的表现。

关键词: P2P网络借贷, 决策树算法, 代价敏感, 非均衡数据集