Imbalanced datasets in the generation of fuzzy classification systems - an investigation using a multiobjective evolutionary algorithm based on decomposition
- 10 November 2016
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 16, 1445-1452
Abstract
In the last years, multi-objective evolutionary algorithms have been used to learn or tune components of fuzzy systems from data. The suitability of such algorithms for this task is due to the possibility of balancing the conflicting objectives of accuracy and interpretability of the resulting model. In a previous work, a method to learn fuzzy classification rules from imbalanced datasets using multi-objective genetic algorithms and the iterative rule learning approach was proposed by the authors. In this method, the imbalanced datasets are pre-processed to be transformed to balanced datasets, and then the rules are generated and the fuzzy sets are tuned. The method has been evaluated in an experimental study considering ten different methods to pre-process the imbalanced datasets and presented competitive results with comparison to similar proposals. The genetic generation of the rules and the optimization of fuzzy sets were both based on NSGA-II. The objective of this article is to investigate whether the multi-objective algorithm used can impact the performance of the method. In this direction, the work developed here presents and analyses the results obtained by the method proposed before using MOEA/D instead of NSGA-II. The analysis demonstrate that the accuracy obtained with MOEA/D is similar to that of NSGA-II while the interpretability measures such as number of rules and number of conditions tend to be better.Keywords
This publication has 37 references indexed in Scilit:
- Designing fuzzy imbalanced classifier based on the subtractive clustering and Genetic Programming2013
- Analysis of preprocessing vs. cost-sensitive learning for imbalanced classification. Open problems on intrinsic data characteristicsExpert Systems with Applications, 2012
- Mine Classification With Imbalanced DataIEEE Geoscience and Remote Sensing Letters, 2009
- Forecasting of ozone episode days by cost-sensitive neural network methodsScience of The Total Environment, 2009
- Multiobjective Optimization Problems With Complicated Pareto Sets, MOEA/D and NSGA-IIIEEE Transactions on Evolutionary Computation, 2008
- Face Recognition Using Total Margin-Based Adaptive Fuzzy Support Vector MachinesIEEE Transactions on Neural Networks, 2007
- Borderline-SMOTE: A New Over-Sampling Method in Imbalanced Data Sets Learning2005
- Class imbalances versus small disjunctsACM SIGKDD Explorations Newsletter, 2004
- Class Imbalances versus Class Overlapping: An Analysis of a Learning System Behavior2004
- Interpretability Issues in Fuzzy Modeling2003