A Meta-Learning Method for Concept Drift

Runxin Wang, Lei Shi, Mícheál Ó Foghlú, Eric Robson

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

The knowledge hidden in evolving data may change with time, this issue is known as concept drift. It often causes a learning system to decrease its prediction accuracy. Most existing techniques apply ensemble methods to improve learning performance on concept drift. In this paper, we propose a novel meta learning approach for this issue and develop a method: Multi-Step Learning (MSL). In our method, a MSL learner is structured in a recursive manner, which contains all the base learners maintained in a hierarchy, ensuring the learned concepts are traceable. We evaluated MSL and two ensemble techniques on three synthetic datasets, which contain a number of drastic concept drifts. The experimental results show that the proposed method generally performs better than the ensemble techniques in terms of prediction accuracy.
Original languageEnglish
Title of host publicationKDIR 2010 - Proceedings of the International Conference on Knowledge Discovery and Information Retrieval
Pages257-262
Number of pages6
Publication statusPublished - 2010
EventInternational Conference on Knowledge Discovery and Information Retrieval, KDIR 2010 - Valencia, Spain
Duration: 25 Oct 201028 Oct 2010

Publication series

NameKDIR 2010 - Proceedings of the International Conference on Knowledge Discovery and Information Retrieval

Conference

ConferenceInternational Conference on Knowledge Discovery and Information Retrieval, KDIR 2010
Country/TerritorySpain
CityValencia
Period25/10/201028/10/2010

Keywords

  • Concept drift
  • Data mining
  • Evolving data
  • Meta-Learning
  • Supervised learning

Fingerprint

Dive into the research topics of 'A Meta-Learning Method for Concept Drift'. Together they form a unique fingerprint.

Cite this