项目名称: 基于支撑向量的生存分析方法的研究与应用
项目编号: No.81202289
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 预防医学、地方病学、职业病学、放射医学
项目作者: 牛晓辉
作者单位: 华中农业大学
项目金额: 23万元
中文摘要: 一方面支持向量机方法较传统的Cox回归模型,具有不需要先验知识、能解决小样本、非线性、过拟合、维数灾难,并且具有推广能力强、全局最优等明显优势;另一方面,传统的支撑向量机方法只能处理完全数据,不能处理包含删失数据的生存数据。 因此,基于支撑向量的生存分析的研究具有重要的理论与实际意义。 本课题首先根据传统的支撑向量机方法,基于结构风险最小化原则,构建能够处理删失数据的数学模型;其次,以常用的三种核函数、构建的新核函数或者修正现有的核函数为候选核函数,从中选取较优者,以提高算法的性能;再次,通过统计模拟数据,人为控制数据的删失比例,通过数值实验,分析不同删失比例下较优的支撑向量机的算法与传统的Cox回归模型的性能,通过综合比较产生最优的算法;最终根据最优的算法,形成一套完整的处理生存数据的方法,并运用于解决两个实际问题,以通过实际问题检验算法的有效性。
中文关键词: 生存分析;支撑向量机;删失数据;删失比例;颅脑损伤
英文摘要: On the one hand, compared with the Cox regression model, the support vector machine(SVM) has many advantages, such as no requirement of priori knowledge, the ability to handle small samples, nonlinear learning, over-fitting, curse of dimensionality, the generalization ability and global optimization. On the other hand, the traditional SVM can not deal with the survival data with the censored data.Therefore, the research of the survival analysis method based on the support vector has important theoretical and practical significance. In this project, the mathematical model for censored data, which is based on the principle of Structural Risk Minimization(SRM), will be firstly constructed according to the traditional SVM. Subsequently, we will try to construct new kernel function, or amend the existing kernel functions. From all of these kernel functions, which include the three commonly used kernel function, the two optimum kernel functions will be chosen to improve the performance of the algorithms. In addition, the candidate algorithms will be composed of the two optimum methods and Cox regression model. Artificially controlling the censored proportion, the statistical simulation data will be constructed. Through the statistical simulation numerical experiments, the relationship between the performance of the c
英文关键词: Survial Analysis;Support vector machine;Censored data;Censored proportion;Traumatic Brain Injuries