Outsourced software project is one of the main ways of software development, which is of high failure rate. Intelligent risk prediction model can help identify high risk project in time. However, the existing models are mostly based on such a hypothesis that all the cost of misclassification is equal, which is not consistent with the reality that in the domain of software project risk prediction, the cost of predicting a fail-prone project as a success-prone project is different from predicting a success-prone project as a fail-prone project. To the best of our knowledge, the cost-sensitive learning method has not yet been applied in the domain of outsourced software project risk management though it has been widely used in a variety of fields. Based on this situation, we selected five classifiers, and introduced cost-sensitive learning method to build intelligent prediction models respectively. This paper totally collected 292 real data of outsourced software project for modeling. Experiment results showed that, under cost-sensitive scenario, the polynomial kernel support vector machine is the best classifier for outsourced software project risk prediction among the five classifiers due to its high prediction accuracy, stability and low cost.