基于概率相关性的多标签数据流变化检测

由于传统的概念漂移检测研究主要针对单标签数据流,对现实中常见的多标签数据流却缺乏足够的关注,多标签数据流概念漂移检测问题有待进一步的研究。因此,通过分析多标签数据流中存在的特殊依赖关系,提出了一种基于概率相关性的多标签数据流概念漂移检测算法。其基本思想是从概念漂移的产生原因出发,利用概率相关性近似描述数据分布来监测新旧数据分布变化,判断概念漂移是否发生。实验结果表明,提出的算法能够比较快速、准确地检测到概念漂移,并在多标签概念漂移数据流分类问题上取得了预期的学习效果。

Detection of Multi-label Data Streams Change Based on Probability of Relevance

Traditional detection approaches of concept drift mainly focus on single-label scenarios,however,not enough attention has been paid to the problem of mining from multi-label data streams.But applications of such data streams are common in the real world.These make it necessary to design efficient algorithms to detect concept drift for multi-label data streams.So after particularly analyzing the unique property label dependence of multi-label data streams,the paper proposed an algorithm of detecting concept drift based on the probability of relevance for multi-label data streams.The basic idea originates from the reason of concept drift and it describes the distribution of data streams by using the probability of relevance.Then,it estimates whether the concept drift occurs or not through monitoring the change of distribution between the old data and new data.The final experimental results show that the proposed algorithm can rapidly and accurately detect the concept drift and achieve prospective predictive performance for multi-label evolving stream classification.