非对称DNA序列混合识别模型研究  

A Research on Mixed Recognition Model for Imbalanced DNA Sequence

作　　者：罗泽举[1] 宋丽红[2] 李艳会[3] 朱思铭[3] 

机构地区：[1]重庆工商大学计算机科学与信息工程学院,重庆400067 [2]重庆工商大学实验实习中心,重庆400067 [3]中山大学数学与计算科学学院,广州510275

出　　处：《计算机科学》2007年第12期207-210,共4页Computer Science

基　　金：重庆市教育委员会科学技术研究项目资助(KJ0707022)

摘　　要：建立了一种改进的不对称支持向量机(MISVM)和隐马尔可夫模型结合的混合学习模型,对于实际中具有的不对称样本数据集,采用调整Hessian矩阵对角参数的策略,增大数据量少的样本离超平面的距离,再结合隐马尔可夫谱变换,以达到更加精确地分离不对称样本的目的。实验发现,不能简单利用正负两类样本所占百分比或固定参数来改变核函数矩阵的对角参数,而必须加之以可调整的权系数才能控制错分的样本数;经改进后的混合不对称学习算法比标准SVM具有更高的分辨率,对启动子序列进行识别,平均识别率达到91.8%。Set up a modified imbalanced SVM（MISVM） mixed learning models associated with HMM, for the imbalanced data set in practice, use the strategy for adjusting the Hessian matrix diagonal parameter,increase the distance between the few samples and the hyperplane, associated with the spectrum transform of hidden markov, this realizes our intention of separating the imbalanced samples more precisely. The experiment indicates that we can＇t simply use the percentage what the positive samples and the negative samples have or fix the diagonal parameter of kernel function, and must add proper weight coefficient which can be adjusted to control the number of error-divided samples, this modified mixed imbalanced learning algorithm obtain higher recognition rate than that of standard SVM, recognize promoter sequence, the average recognition rate come to 91.8%.

关 键 词：非对称DNA序列 核参数 隐马尔可夫谱 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]