基于遗传算法参数优化的最小二乘支持向量机财务困境预测  

Financial Distress Forecast of Least Squares Support Vector Machines Based on Genetic Algorithm Parameter Optimization

作　　者：赵冠华[1] 李玥[1] 赵娟[1] 

机构地区：[1]山东财政学院,济南250014

出　　处：《科学与管理》2011年第5期56-63,共8页Science and Management

基　　金：山东省科技攻关计划资助项目(2008GG30009005);2008.10-2010.10;山东省软科学研究计划资助项目(2008RKA223);2008.6-2010.6

摘　　要：传统支持向量机应用于财务困境预测时,需要求解复杂的二次规划问题,求解难度大。而最小二乘支持向量机模型可以将二次规划问题变成一个线性方程组来求解,有效降低了模型求解的难度。尤其是将遗传算法应用于最小二乘支持向量机模型参数和核参数的优化时,显著提高了模型预测的正确率。本文从沪深两市随机抽取了2002年-2007年252家A股上市公司作为研究样本,并把研究样本分为两组,对这两组样本数据分别进行了短期及中长期预测。实证结果表明,基于遗传算法的最小二乘支持向量机模型的预测效果不但好于传统统计类Logit模型,也优于传统支持向量机模型。短期预测效果显著优于中长期预测效果,训练样本数直接影响到模型的预测效果,二者呈正相关关系。When using traditional support vector machine to make financial distress prediction, we need to solve the complex quadratic programming problems, which are quite difficult. At the same time, the least squares support vector machine （LS-SVM） can solve the quadratic programming problems by transferring them into linear equations, effectively reducing the difficulty. Especially when applying genetic algorithm to optimize parameters and kernel parameters of LSSVM, the prediction accuracy is significantly improved. We randomly selected 252 A-share listed companies during 2002-2007 from Shanghai and Shenzhen Stock Exchanges as the research samples and divided them into two Sample I and Sample II. Then we carried out short-term and long-term predictions of these two sets of samples res The empirical results showed that the prediction effects of LS-SVM model based on genetic algorithm was better of traditional statistical Logit Model as well as the traditional support vector machine. higher accuracy rate compared with long-term prediction. groups - pectively. than that Besides, short-term prediction had a In addition, the number of training samples directly affected the prediction accuracy and they were positively correlated.

关 键 词：遗传算法 最小二乘支持向量机 参数优化 短期预测 中长期预测 

分 类 号：F830.59[经济管理—金融学]