对于大量的非规范化的临床医学数据来说，传统的统计方法存在一定的局限性，容易造成误判，将数据挖掘技术应用到临床医学数据分析中，逐渐成为热点研究课题。支持向量机(SVM)是近年来机器学习领域中较为高效的工具，支持向量机以统计学习理论为理论基础，求解分类和回归问题，并将问题转化为一个二次规划问题。

支持向量机的凸二次规划问题是一类特殊的约束最优化问题，对于凸规划问题来说局部最优解和全局最优解是等价的。对于二分类的线性可分数据集，考虑求得最大间隔分离超平面，得到基本支持向量机学习的最优化问题。结合最优化理论与方法，将支持向量机中的凸二次规划原始问题转换成对偶问题，应用拉格朗日对偶性，通过求解对偶问题得到原始问题的最优解。为解决近似线性可分数据集的二分类问题，引入松弛变量和惩罚参数，得到软间隔支持向量机数学规划模型。通过核技巧实现到高维空间的非线性映射，用核函数代替对偶问题的目标函数中输入实例与实例之间的内积，构造核函数以及多核函数支持向量机，适合于解决本质上非线性的分类、回归等问题。本文对 GVPM 算法进行数值实验，在此基础上对算法进行了改进，并证明了改进算法的收敛性。改进算法的数值实验中迭代次数显著降低、迭代时间明显缩短。

最后，针对骨科患者的生物力学特征的临床数据，在选择二次多项式核函数的基础上，应用改进的 GVPM 算法，在训练集上训练模型，得到了分离超平面的法向量与截距，进而得到分类决策函数的表达式；对正常患者与异常患者的分类准确率约为 85.16%，对异常患者中椎间盘突出与脊柱滑脱的分类准确率约为 96.82%。

For a large number of non-standard clinical medical data, traditional statistical methods have certain limitations, easy to cause misjudgment, the application of data mining technology to clinical medical data analysis, has gradually become a hot research topic. SVM is a relatively efficient tool in the field of machine learning in recent years. Based on statistical learning theory, SVM solves classification and regression problems, and transforms the problems into quadratic programming problems.

Convex quadratic programming of SVM is a special constrained optimization problem for which local and global optimal solutions are equivalent. For binary linear fractional data sets, the maximum interval separation hyperplane is considered, and the optimization problem of basic support vector machine learning is obtained. Combined with optimization theory and method, the original convex quadratic programming problem in support vector machine is transformed nto a dual problem, and the optimal solution of the original problem is obtained by solving the dual problem using Lagrange duality. In order to solve the binary classification problem of approximately linear fractional data sets, a mathematical programming model of soft interval support vector machines is developed by introducing relaxation variables and penalty parameters. The kernel technique is used to realize the nonlinear mapping to high dimensional space, and the kernel function is used to replace the inner product between the input instances in the objective function of the dual problem. The kernel function and multi-kernel function support vector machine are constructed, which is suitable for solving the classification and regression problems which are nonlinear in nature. In this paper, the numerical experiment of GVPM algorithm is carried out, on the basis of which the algorithm is improved, and the convergence of the improved algorithm is proved. In the numerical experiment of the improved algorithm, the iteration times and iteration time are significantly reduced.

Finally, according to the clinical data of biomechanical characteristics of orthopedic patients, on the basis of selecting quadratic polynomial kernel function, the improved GVPM algorithm was applied to train the model on the training set, and the normal vector and intercept of the separation hyperplane were obtained, and then the expression of the classification decision function was obtained. The classification accuracy of normal and abnormal patients was 85.16%, and the classification accuracy of abnormal disc hernia and spondylolisthesis was 96.82%.