Abstract: Aiming at overcoming the blindness and subjectivity of selecting indicators and confirming indicators’ weight without any experience，an intelligent method of software project risk assessment based on feature evaluation and least squares support vector machine is proposed.By estimating the capacity of each traditional indicator in evaluating software project risk，the sensitive features can be selected and input into support vector machine to identify different conditions of project risk automatically.The results of the demonstration show that the method can enhance the system’s robustness，and select sensitive ones from a large number of software project risk-rating indicators to assess software project risk more correctly.

Key words: feature evaluation, least squares support vector machine, software project, risk assessment

摘要： 为了克服在无先验知识的情况下，人为选择评价指标的盲目性和确定指标权重的主观性，提出一种基于特征评判与最小二乘支持向量机的软件项目风险智能评价方法。该方法对各传统软件风险评价指标进行特征评判，并根据评判因子的大小选取敏感指标作为支持向量机的输入，实现对不同风险状态的自动识别。实证结果表明，该方法具有很强的鲁棒性，能够从大量的软件项目风险评价指标中筛选出有效的敏感特征，准确评价软件项目风险。

关键词: 特征评判, 最小二乘支持向量机, 软件项目, 风险评价