Hybrid Classifier Ensemble for Imbalanced Data.

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Class imbalance is a significant challenge in machine learning.
• Existing methods like undersampling and cost-sensitive learning have limitations, including information loss and sensitivity to noise.
• There is a need for robust methods to handle imbalanced data effectively.

Purpose of the Study:

• To propose a hybrid optimal ensemble classifier framework to overcome limitations of conventional imbalance learning methods.
• To combine density-based undersampling and cost-sensitive classification using multi-objective optimization.
• To improve classification accuracy on imbalanced datasets.

Main Methods:

• Developed a density-based undersampling method with probability-based data transformation to create balanced subsets.
• Employed a cost-sensitive classification approach by adjusting weights of misclassified minority samples.
• Integrated a multi-objective optimization procedure within an ensemble classifier framework to refine classification results.

Main Results:

• The proposed hybrid optimal ensemble classifier demonstrated superior performance compared to existing imbalance and ensemble classification methods.
• Experiments on real-world datasets validated the effectiveness of the combined approach.
• The method successfully addressed information loss and noise sensitivity issues.

Conclusions:

• The hybrid optimal ensemble classifier framework offers a robust solution for class imbalance problems.
• The integration of density-based undersampling, cost-sensitive learning, and multi-objective optimization is effective.
• This approach enhances classification performance on imbalanced datasets.