Online Peer-to-Peer (P2P) lending market has experienced a period of rocketing development far beyond our expectation, which is also facing more challenges, such as the default on a loan. The study aims to explore a data-driven approach to extract knowledge of default risk from borrowers’ demographic and the behavior characteristics in the loaning process, which can be used to reduce the default risk of P2P platform. The possibility of credit risk rating automation can also be investigated by estimating the predicting accuracy. A huge dataset from a famous P2P lending platform in China was analyzed, and three default prediction models were employed for the data research of discrete input-output pairs, continuous input-output pairs and continuous input and discrete output pairs. The average percent hit rate (APHR) and the lift analysis were adopted to evaluate the predictive accuracy. A 2-layer artificial neural network (ANN) model has performed brilliantly with continuous input-output data pairs with an average relative-error value of 0.24. The support vector machine (SVM) is highly recognized due to a predictive accuracy of 89.18% with discrete input-output data pairs. Decision tree C5.0 (DT) was utilized to find some important factors affecting risk rate and predict the default risk of the borrower. The behavior data such as delinquency history, the already paid installment and loan remaining was indispensable on obtaining a higher predictive accuracy. Some constructive conclusions on risk management of P2P online lending can be drawn based on data mining results.