Abstract: In order to improve the accuracy and efficiency of data flow classification detection, a new Gauss naive Bayes classification method based on weighted mechanism concept drift detection is proposed. Firstly, the proposed algorithm framework is designed, and the input data stream is used to establish the information table directly, and the Gauss naive Bayes classifier based on the information table is also constructed; Secondly, the Kappa statistical method is used to establish the concept drift detection method. According to the input data fluctuation, linear function and Bias function (nonlinear) are taken to detect the concept drift, and expert point deletion and information table are used to deal with the recurrent concept drift, to improve the drift detection accuracy and efficiency; Finally, simulation experiments show that the classification accuracy on the SEA test set, Hyperplane data set and SQD data set is 10.3 %, 16.8 % and 20.5 % higher than that of the contrast algorithm, which verifies the effectiveness of the classification algorithm.

Key words: Weighted mechanism, concept drift, data flow, Gauss, naive Bayes

摘要： 为提高数据流分类检测精度和检测效率，提出一种基于加权机制概念漂移策略的数据流高斯朴素贝叶斯分类检测算法。首先，对所提算法框架进行设计，利用输入数据流直接建立信息表，并构建基于信息表的高斯朴素贝叶斯分类器；其次，利用“Kappa统计”方法建立基于加权机制的概念漂移检测方法，根据输入数据波动性，分别采取线性函数和贝叶斯（非线性）函数进行检测，并利用专家点删除和信息表来处理经常性的概念漂移，实现漂移检测精度和效率的提升；最后，通过仿真实验，显示所提算法在SEA测试集、Hyperplane数据集和SQD测试集上的分类精度分别比选取的对比算法提高分类精度10.3 %、16.8 %和20.5 %以上，验证了所用分类检测算法的有效性。

关键词: 加权机制, 概念漂移, 数据流, 高斯, 朴素贝叶斯