Unsupervised Concept Drift Detection in Data Streams

In data stream mining, efficiently detecting concept drifts is still challenging due to the high cost of collecting true class labels. Traditional detection methods usually need high computation and memory cost and is unable to distinguish between concept drift and novelty. To improve the drift detection efficiency, we propose four unsupervised concept drift detection methods. Especially, we propose Radial Distance Drift Detection (RDDD) which aims to improve drift detection efficiency by allowing adjustable sensitivity to changes in different feature dimensions. The method Novelty-aware Concept Drift Detection (NACD) is designed to simultaneously detect both concept drift and data novelty. In summary, approaches taken in this thesis are novel and practical. There has been no previous work on drift detection with adjustable feature sensitivity or the ability to distinguish drift from novelty.

Author dc:creator

• Shang, Dan

Statement dc:rights

• info:eu-repo/semantics/openAccess
• The author owns the copyright in this thesis including all reproduction and reuse rights for the work. The work may not be altered without the permission of the copyright owner. Attribution is essential when quoting or paraphrasing from this thesis.
• © 2024 Dan Shang
• au.edu.uts.lib/cph

Language dc:language.iso

en_US

Handle dc:identifier.uri

http://hdl.handle.net/10453/187954

OAI identifier oai:identifier

oai:opus.lib.uts.edu.au:10453/187954