基于高斯Copula的约束贝叶斯网络分类器研究

发布时间：2019-03-14 13:51

【摘要】：具有连续属性的分类问题普遍存在,目前主要采用两种方法来处理连续属性:一种是将连续属性进行离散化;另一种是基于高斯函数或高斯核函数来估计属性密度.连续属性的离散化可能导致信息丢失、引入噪声和类对属性的变化不够敏感等问题,而高斯函数和高斯核函数在属性密度估计中各有优势与不足,但它们具有很强的互补性.该文依据Copula和贝叶斯网络理论,结合高斯Copula密度函数、引入平滑参数的高斯核函数和以分类准确性为标准的属性父结点贪婪选择,建立连续属性约束贝叶斯网络分类器,既可以避免连续属性离散化所带来的问题,又能够实现高斯函数和高斯核函数在属性密度估计方面的优势互补.分别采用真实数据和模拟数据进行实验,结果显示,使用结合边缘高斯核函数的高斯Copula估计属性密度的约束贝叶斯网络分类器具有良好的分类准确性.
[Abstract]:At present, there are two main methods to deal with continuous attributes: one is to discretize the continuous attributes; the other is to estimate the density of attributes based on Gao Si function or Gao Si kernel function. Discretization of continuous attributes may lead to loss of information, the introduction of noise and the insensitivity of classes to the change of attributes, while Gao Si's function and Gao Si's kernel function have their own advantages and disadvantages in the estimation of attribute density, but they are complementary to each other. According to the theory of Copula and Bayesian network, combining the density function of Gao Si Copula, this paper introduces Gao Si kernel function of smooth parameter and greedy selection of attribute parent node based on classification accuracy, and establishes continuous attribute constrained Bayesian network classifier. It can not only avoid the problems caused by discretization of continuous attributes, but also realize the complementary advantages of Gao Si's function and Gao Si's kernel function in the estimation of attribute density. Experiments with real data and simulated data show that the constrained Bayesian network classifier using Gao Si Copula combined with the marginal Gao Si kernel function to estimate the density of attributes has good classification accuracy.
【作者单位】：上海立信会计学院数学与信息学院;上海立信会计学院立信会计研究院;
【基金】：国家自然科学基金(61272209) 上海市自然科学基金(15ZR1429700) 上海市教委科研创新项目(15ZZ099)资助
【分类号】：TP181

【相似文献】