We present an ensemble method for concept drift that dynamically createsand removes weighted experts in response to changes in performance.The method, dynamic weighted majority (DWM), uses four mechanismsto cope with concept drift:It trains online learners of the ensemble, it weights those learnersbased on their performance, it removes them, also based on theirperformance, and it adds new experts based on the global performanceof the ensemble.After an extensive evaluation---consisting of five experiments,eight learners,and thirty data sets that varied in type of targetconcept, size, presence of noise, and the like---we concluded thatDWM outperformed other learnersthat only incrementally learn concept descriptions,that maintain and use previously encountered examples, andthat employ an unweighted, fixed-size ensemble of experts. color="gray">
展开▼
机译：我们提出了一种概念漂移的整体方法，该方法可以动态地创建和删除性能变化的加权专家。该方法称为动态加权多数（ DWM ），它使用四种机制来应对概念漂移：它培训在线学习者对整个集成进行评估，根据学习者的表现对他们进行加权，然后根据他们的表现将其删除，然后根据总体的学习表现添加新专家。经过广泛的评估-由五个实验，八个学习者和在目标概念的类型，大小，噪声的存在等方面有所变化的三十个数据集—我们得出的结论是， DWM 优于仅学习概念描述，保持并使用先前遇到的示例的其他学习者，并且聘用了不加权，固定大小的专家团队。 color =“ gray”>
展开▼
