P2P网贷平台的高速发展, 降低了小微企业的借贷成本, 提高了投资者的收益与效率, 较好地满足了小微企业的融资需求.然而, 现阶段中国的P2P网贷平台在发展过程中也暴露出大量的风险问题, 不仅使投资者财富遭受损失, 也严重危害了P2P行业的健康发展.因此, 对P2P网贷平台进行早期 风险预测, 在风险问题未发生之前对投资者进行风险预警并为投资者提供投资辅助决策是目前学术界广受关注的一个热点研究问题.针对真实P2P网贷平台数据的类别分布非均衡性问题, 文章提出了一种基于$K$-Means聚类和支持向量机(support vector machine, SVM)的非均衡分类方法SVM${}^{K\text{-}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{s}}$用以预测P2P网贷平台风险.通过网贷之家真实数据 并以经典的逻辑回归(logistic regression)、支持向量机以及神经网络(back propagation neural network)为基准方法进行的比较试验表明, 文章提出的SVM${}^{K\text{-}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{s}}$方法能够更加准 确地在早期进行P2P网贷平台风险预测.

The rapid development of P2P online loan platform reduces the lending cost of startup enterprises and improves profit and return of investors. However, the development of P2P lending platforms in China has exposed a large number of risk problems, which not only hurt investors' wealth, but also seriously endangers the healthy development of P2P industry. Therefore, early risk prediction of P2P lending platform before bursting of loan risks to support investors in decision making on investment is currently a hot problem in the academia research cycle. In most cases, the data from P2P lending platforms is unbalanced, i.e., the number of defrauding loans is small while the number of non-defrauding loans is large. With the real data collected from the WangDaiZhiJia website, this paper proposes a novel approach called SVM${}^{K\text{-}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{s}}$ for unbalanced classification problem to predict the early risks of those P2P lending platforms. This paper also uses classic logistic regression, support vector machine and back propagation neural network as the baseline methods for performance comparison. The experimental result shows that the proposed SVM${}^{K\text{-}\mathrm{M}\mathrm{e}\mathrm{a}\mathrm{n}\mathrm{s}}$ approach performs better than the baseline methods on early risk prediction of P2P lending platforms.