提出一种基于粗糙集与支持向量机的客户动态评估方法。根据客户群特点从当前价值、潜在价值和附加价值三个维度分析并构建客户评估指标，利用指标的年增幅率监测客户价值的变化规律。应用粗糙集布尔推理算法、粒子群算法实现连续属性离散化和知识约简。通过10-重交叉验证和网格搜索技术获取最优惩罚因子与核参数，缩放样本数据集并完成支持向量机一对一分类器的训练与测试。结果表明该评估方法能够实现周期性的客户价值评估与细分，具有很强的泛化能力。- A customer dynamic evaluation method based on rough set and support vector machine is advanced. Customer evaluation indicators are analysed and established from three dimensions of current value, potential value and odditional value according to customer characteristics. The change rules of customer value are observed by annual increasing rate of indicators. Continuous attributes are discretized by rough set and boolean inference arithmetic. Redundant attributes are reduced by particle swarm optimization arithmetic. The optimal penalty factor and nuclear parameter are obtained by 10-cross validation and grid-search. The sample data scaling is carried out and the train and test of svm one against one classifier are accomplished. The result indicates that the evaluation method can not only implement the periodic evaluation and classification of customer value, but also have a better generalization.