基于三对角矩阵的完全贝叶斯分类器研究  

Research of full Bayesian classifiers based on tridiagonal matrix

作　　者：冷翠平[1] 王双成[1,2] 杜瑞杰[2] 

机构地区：[1]上海立信会计学院数学与信息学院,上海201620 [2]上海立信会计学院开放经济与贸易研究中心,上海201620

出　　处：《计算机应用研究》2015年第3期740-742,747,共4页Application Research of Computers

基　　金：国家教育部人文社科基金资助项目(10YJA630154);上海市教委科研创新项目(11YZ240);上海市教委重点学科(第五期)建设项目(J51702)

摘　　要：针对连续属性朴素贝叶斯分类器不能有效利用属性之间的条件依赖信息、而对其进行依赖扩展中的高阶协方差矩阵的求逆和行列式运算又非常困难等问题,将三对角矩阵和多元高斯函数相结合,建立连续属性完全贝叶斯分类器,并在三对角矩阵中引入平滑参数,通过对平滑参数的调整来实现分类器的优化。使用UCI数据的实验结果显示,经过优化的连续属性完全贝叶斯分类器具有良好的分类准确性。Naive Bayes classifiers with continuous attributes can not make full used of dependency information between attrib- utes. And in their dependency extension, it is difficult to calculate the inverse and the determinant of the high-order covari- ance matrix. To solve the dilemma, combining the tridiagonal matrix and multivariate Gaussian function can help build the full Bayesian classifiers with continuous attributesl And introduced smoothing parameters to tridiagonal matrix, which can contrib- ute to the classifiers optimization. Experimental results resulting from using UCI data show that the optimized classifiers with continuous attributes have good classification accuracy.

关 键 词：朴素贝叶斯分类器 完全贝叶斯分类器 多元高斯函数 三对角矩阵 平滑参数 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]