【摘要】 通过对数据流分类中的概念漂移问题的研究,提出了一种在线装袋(Online Bagging)算法的改进算法——自适应抽样参数的在线装袋算法AdBagging(adaptive lambda bagging)。利用在分类过程中出现的误分样本数量来调整Online Bagging算法中的泊松(Poisson)分布的抽样参数,从而可以动态调整新样本在学习器中的权重,即对于数据流中的误分类样本给予较高的学习权重因子,而对于正确分类的样本给予较低的学习权重因子,同时结合样本出现的时间顺序调整权重因子,使得集成分类器可以动态调整其多样性(adversity)。该算法具有OnlineBagging算法的高效简洁优点,并能解决数据流中具有概念漂移的问题,人工数据集和实际数据集上的实验结果表明了该算法的有效性。更多还原

【Abstract】 By analyzing the concept drift problem in data stream classification,a new algorithm based on online bagging algorithm named adaptive lambda bagging algorithm(AdBagging) is introduced.The new algorithm dynamically adjusts the sampling parameters of Poisson distribution in online bagging algorithm based on the number of misclassified samples in data stream classification.Through this procedure the new algorithm could give more attention to the misclassified samples and give little attention to the right classified samples.At the same time the learning weight of samples for algorithm can be adjusted according to the temporal order.So the algorithm could solve the concept drift problem in data stream classification.Experiments on synthesize and real data sets prove that the algorithm is effective.更多还原