Multi-label classification via incremental clustering on an evolving data stream

作者：

Highlights：

• An incremental clustering-based multi-label online classification algorithm for multi-label data stream is proposed.

• To handle concept drift, our algorithm evolves with time, giving higher attention to more recent samples than older samples through a weight decay mechanism.

• Our algorithm dynamically determines the number of predicted labels based on Hoeffding inequality and the label cardinality.

• Extensive comparative experiments with the state-of-the-art algorithms validated the superior performance of our algorithm in both the stationary and concept drift settings.

摘要

•An incremental clustering-based multi-label online classification algorithm for multi-label data stream is proposed.•To handle concept drift, our algorithm evolves with time, giving higher attention to more recent samples than older samples through a weight decay mechanism.•Our algorithm dynamically determines the number of predicted labels based on Hoeffding inequality and the label cardinality.•Extensive comparative experiments with the state-of-the-art algorithms validated the superior performance of our algorithm in both the stationary and concept drift settings.

论文关键词：Multi-label classification,Incremental learning,Online learning,Clustering,Data stream,Concept drift

论文评审过程：Received 31 October 2018, Revised 5 March 2019, Accepted 1 June 2019, Available online 3 June 2019, Version of Record 11 June 2019.

论文官网地址：https://doi.org/10.1016/j.patcog.2019.06.001