一种多分辨率数据流在线分类算法

Multi-resolution online classification algorithm for data streams

提出一种能够适应数据流突变式概念变化的增量分类算法,采用网格技术对数据集特征向量进行量化,利用Haar小波多种分辨率的数据表示方式,基于最近邻技术发现测试点的合适类标签。在真实数据集上的测试证明,与已存在的数据流分类算法相比,提出的分类算法精度较高,具有很低的更新代价,适合数据流应用的需求。

An incremental classification algorithm based on nearest neighbor technology, which adapts to the sudden concept shift over data streams, was proposed. The algorithms uses grid technique to quantize the feature space of data set and uses a muhi-resolution data representation based on Hanr wavelets to find adaptive class label of a test point. Experiments performed on both synthetic and real-life data indicate that the proposed classifier outperforms existing algorithms for data streams in terms of accuracy, The algorithm ＂s low update and computational cost makes it highly suitable for data stream applications.