论文部分内容阅读
文章提出了一种新的贝叶斯分类算法——加权灵活贝叶斯分类器用以处理连续值属性的分类问题。通过度量条件属性与决策属性的相关度,类条件概率中对应的边缘概率密度被赋予了相应的权重,其中,条件属性与决策属性之间的相关性通过基于互信息的相关度量标准来计算。在10个UCI数据集合上面,我们比较了WFNB与加权朴素贝叶斯和灵活贝叶斯分类器的分类精度,试验结果表明,我们提出的WFNB有效地改进了传统贝叶斯分类器的分类精度。
In this paper, a new Bayesian classification algorithm - weighted flexible Bayesian classifier is proposed to deal with the classification of continuous-valued attributes. By measuring the correlation between the condition attributes and the decision attributes, the corresponding edge probability density in the condition-like probability of the class is given a corresponding weight, and the correlation between the condition attributes and the decision attributes is calculated by the related metric based on the mutual information. In the 10 UCI data sets above, we compare the classification accuracy of WFNB with Weighted Naive Bayes classifier and flexible Bayes classifier. The experimental results show that our proposed WFNB effectively improves the classification accuracy of traditional Bayesian classifier .