论文部分内容阅读
文中首先分析降噪集成算法采用的样本置信度度量函数的性质,阐述此函数不适合处理多类问题的根源.进而设计更有针对性的置信度度量函数,并基于此函数提出一种增强型降噪参数集成算法.从而使鉴别式贝叶斯网络参数学习算法不但有效地抑止噪声影响,而且避免分类器的过度拟合,进一步拓展采用集群式学习算法的鉴别式贝叶斯网络分类器在多类问题上的应用.最后,实验结果及其统计假设检验分析充分验证此算法比目前的集群式贝叶斯网络参数学习方法得到的分类器在性能上有较显著提高.
In this paper, we first analyze the properties of the sample confidence measure function used in the noise reduction integrated algorithm, and explain that this function is not suitable for dealing with the root causes of many kinds of problems. Then we design a more targeted confidence measure function and propose an enhanced So that the discriminant Bayesian network parameter learning algorithm not only effectively restrain the influence of noise but also avoids the over fitting of the classifier and further expands the discriminant Bayesian network classifier using the cluster learning algorithm Finally, the experimental results and its statistical hypothesis test and analysis fully verify that the proposed algorithm has a better performance than the current class-based Bayesian network parameter learning method.