Abstract: Aiming at the phenomenon that imputation methods to deal with incomplete data will introduce additional noise, this paper proposes a novel forecast model based on max-margin principle. This method directly uses data with absent features to predict, and experimental results show that compared with imputation methods, this model is more effective in the prediction accuracy.

Key words: max-margin principle, incomplete data, software reliability, reliability prediction

摘要： 针对现有处理不完全数据的填充方法，对数据集引入新的噪声这一问题，提出一种基于最大间隔理论的预测模型，直接使用含缺失特征的样本进行预测，通过实验验证该模型优于常用的填充模型。

关键词: 最大间隔, 不完全数据, 软件可靠性, 可靠性预测