知识驱动的支持向量机理论、算法与应用研究

项目名称： 知识驱动的支持向量机理论、算法与应用研究

项目编号： No.11271361

项目类型： 面上项目

立项/批准年度： 2013

项目学科： 数理科学和化学

项目作者： 田英杰

作者单位： 中国科学院大学

项目金额： 50万元

中文摘要： 本项目主要研究数据挖掘中的最优化方法- - -支持向量机的新理论、新算法及其应用。我们将从最优化理论与方法的角度，研究知识驱动的支持向量机，主要包括三个方面：一是近年来机器学习领域的研究热点：非平行超平面支持向量机，与标准的支持向量机相比，该方法更灵活，推广能力更强。我们将研究其统计学习理论基础，大规模优化问题的高效求解方法，以及其拓展的优化模型与方法等；二是基于先验知识的支持向量机，研究如何将某些应用领域的专家经验或知识与SVM结合，寻求更加有效的最优化模型和方法；三是基于支持向量机的知识提取，研究如何使支持向量机这一"黑箱"模型透明化，学习结果规则化与知识化，特别是寻求和建立相应的最优化模型和方法。这三方面以非平行超平面支持向量机理论与算法为基础，把知识与支持向量机有机地结合在一起。关于应用，我们将利用上述理论研究成果，以国内某商业银行的信用卡客户数据为基础，进行信用卡流失分析方面的研究。

中文关键词： 支持向量机；分类；知识驱动；非平行超平面；数据挖掘

英文摘要： This research proposal focuses on the optimization method in Data Mining- - Support Vector Machines (SVMs) New Theory, Algorithms and Applications. We will research on the knowledge-driven support vetor machines from the perspective of optimization theory and method, mainly including the following three parts: The first is the nonparallel hyperplane support vector machines, recent research hotspot in the field of machine learning. Compared with the standard SVMs, nonparallel hyperplane SVMs are more flexible and with higher generalization ability. We will research their statistical learning theory foundation, the efficient solving methods for the large scale optimization problems, their extended optimization models and methods and etc; The second is the knowledge-based support vector machines. We will research on the methods combining the expert experiences or knowledge of some application fileds with SVMs, to search for more efficient optimization models and methods; The third is the rule extraction from SVMs. How to make SVMs more transparent, and the learning results to be more explainable and understandable rules or knowledge are under our consideration, especially searching for the optimization models and methods. The above three parts combine SVMs with knowledge sucessfully based on the theory and algor

英文关键词： support vector machines；classification；knowledge driven；nonparallel hyperplane；data mining