Modifications of classification strategies in rule set based bagging for imbalanced data

Authors:
Krystyna Napierala;Jerzy Stefanowski
Affiliations:
Institute of Computing Science, Poznań University of Technology, Poznań, Poland;Institute of Computing Science, Poznań University of Technology, Poznań, Poland
Venue:
HAIS'12 Proceedings of the 7th international conference on Hybrid Artificial Intelligent Systems - Volume Part II
Year:
2012

Citing 8
Cited 0

Rule induction with CN2: some recent improvements

EWSL-91 Proceedings of the European working session on learning on Machine learning
Bagging predictors

Machine Learning
A simple, fast, and effective rule learner

AAAI '99/IAAI '99 Proceedings of the sixteenth national conference on Artificial intelligence and the eleventh Innovative applications of artificial intelligence conference innovative applications of artificial intelligence
Towards tight bounds for rule learning

ICML '04 Proceedings of the twenty-first international conference on Machine learning
Learning from Imbalanced Data

IEEE Transactions on Knowledge and Data Engineering
Ensembles of Abstaining Classifiers Based on Rule Sets

ISMIS '09 Proceedings of the 18th International Symposium on Foundations of Intelligent Systems
Improved heterogeneous distance functions

Journal of Artificial Intelligence Research
On combined classifiers, rule induction and rough sets

Transactions on rough sets VI

Quantified Score

Hi-index	0.00

Visualization

Abstract

Learning bagging ensembles of rule classifiers from imbalanced data is considered. We claim that simply introducing bagging instead of single classifiers may not bring the expected improvement in recognizing a minority class. The reason lies in the classification strategies of component classifiers, which are biased toward majority classes when no-matching or multiple-matching conflicts between rules occur. We argue that abstaining, i.e. allowing component classifiers to refrain from giving a prediction in ambiguous situations, may help to correctly recognize minority examples. Our evaluation on 17 imbalanced datasets and 5 classification strategies shows that bagging with abstaining is better than both standard bagging and single rule based classifiers.