中文摘要：

数据流环境下的高维、属性冗余、含噪音等问题是经常且可能同时存在的,在一定程度上影响了数据流的分类效果.为改善这一现状,提出一种快速、有效的数据流特征选择方法.引入统计指标IV（information value）值作为特征重要度的评价标准,在此基础上依据经验阈值来进行特征选择,从而解决了传统特征选择方法时空效率不高、区分度不明显、难以应用数据流的问题.实验结果表明：FS-IV具有较小的时间开销和较好的抗噪性能,该方法与已有的数据流分类模型相结合,在保证分类精度可比的情况下,能显著提高时空性能.

英文摘要：

The problems of high-dimension, redundant features, and noise, which exist usually and simultaneously in the data stream, lead to long training time and low classification accuracy. An effective and real-time feature selection approach （feature selection-information value, FS-IV） was proposed for the data stream classification. In the FS-IV approach, a statistical index, the information value （IV）, was introduced to measure the importance of features, and the feature was selected according to the threshold of IV value. Therefore, the FS-IV overcomes the problem of expensive cost of time and space and the problem of unobvious distinguishability in classical feature selection approaches in data stream. The experimental result shows that the FS-IV approach is little-cost and anti-noisy, and the FS-IV combined with the data stream classification approach can perform with notable shortened time while maintaining the accuracy.