基于动态约简的增量贝叶斯分类算法的研究

ON INCREMENTAL NAVE BAYESIAN CLASSIFICATION ALGORITHM BASED ON DYNAMIC REDUCTION

朴素贝叶斯由于条件独立性假设使其分类效果不明显,同时在处理海量数据时缺乏灵活性。针对以上情况,提出一种基于动态约简的增量贝叶斯分类算法。算法首先利用(F-λ)广义动态约简计算出数据集的核属性,然后根据训练集的先验信息构造分类器对测试实例进行分类,最后利用类置信度进行选择性增量学习,增强处理增量数据的能力。实验结果表明,该算法在处理属性少的小量数据时,分类效果有一定的改善,在处理多属性大量数据时,分类效果明显提高。

The classification effect of nave Bayesian is not obvious because of conditional independence assumption,so does its lack of flexibility in dealing with massive data. In view of the above,we propose a dynamic reduction-based incremental nave Bayesian classification algorithm. This algorithm uses（ F-λ） generalised dynamic reduction to calculate the core attributes of dataset first,and then constructs classifier according to priori information of training sets to classify the test cases. Finally, it uses class confidence to conduct selective incrementallearning for enhancing the capability of incremental data processing. Experimental results show that the algorithm ameliorates the classification effect to a certain extent when dealing with few attributes and small amount of data,and when dealing with more attributes and large amount of data,the classification effect is obviously improved as well.