基于样本不确定性的增量式数据流分类研究

Research of Incremental Data Stream Classification Based on Sample Uncertainty

具有概念漂移的数据流分类应用场景逐渐增多,如何解决该类问题成为研究热点.文中根据数据流概念漂移特征,结合增量学习原理实现基于样本不确定性选择策略的增量式数据流分类(IDSCBUC)模型.分类模型用支持向量机作为训练器,基于当前分类器从相邻训练集中按照样本不确定性值选择出"富信息"样本代表新概念样本集,把新概念样本集与支持向量集合并更新分类器,形成新的分类模型.理论分析和实验结果表明该方案是可行的,且具备抗噪声能力.

Application of data stream classification with concept drift is gradually increasing, how to solve the problem is becoming re- search hot. According to the traits of concept drift and the principle of incremental learning, incremental data stream classification based sample uncertainty （IDSCBUC） is presented in this paper. Based on the current classifier, new concept sample set are chose by sample uncertainty,and then support vector machine is used to train new classifier model with support vector set and new concept sample set. Through theoretical analysis and simulation experiment,it shows IDSCBSU is feasible and owns anti-noise ability.