Abstract:

In data stream ensemble classification， to make the classifiers adapt to the constantly changing data stream and adjust the weights of base classifiers to select an appropriate set of classifiers， an ensemble classification algorithm based on dynamic weighting function was proposed. Firstly， a new weighting function was proposed to adjust the weights of the base classifiers， and the classifiers were trained with constantly updated data blocks. Then a weight function was used to make a reasonable selection of candidate classifiers. Finally， the incremental nature of decision tree was applied to the base classifiers， and the classification of data stream was realized. Through a large amount of experiments， it is found that the performance of the proposed algorithm is not affected by block size. Compared with AUE2 algorithm， the average number of leaves is reduced by 681.3， the average number of nodes is reduced by 1 192.8， and the average depth of the tree is reduced by 4.42. At the same time， the accuracy is relatively improved and the time-consuming is reduced. Experimental results show that the algorithm can not only guarantee the accuracy but also save a lot of memory and time when classifying data stream.

Key words: data stream, ensemble classification, dynamic weighting, block, incremental learning

摘要：

针对数据流集成分类如何使分类器适应不断变化的数据流，调整基分类器的权重选择合适的分类器集合的问题，提出了一种基于动态加权函数的集成分类算法。首先，提出了一种加权函数调节基分类器的权重，使用不断更新的数据块训练分类器；然后，使用一个新的权重函数对候选分类器进行一个合理的选择；最后，在基分类器中应用决策树的增量性质，实现对数据流的分类。通过大量实验发现，基于动态加权函数的集成分类算法的性能不受块的大小影响，与AUE2算法相比，叶子数平均减少了681.3、节点数平均减少了1 192.8，树的深度平均减少了4.42，同时相对地提高了准确率，降低了消耗时间。实验结果表明该算法在对数据流进行分类时不但可以保证准确率还可以节省大量的内存空间和时间。

关键词: 数据流, 集成分类, 动态加权, 块, 增量学习