Experimental Study and Comparison of Imbalance Ensemble Classifiers with Dynamic Selection Strategy.

Area of Science:

• Machine Learning
• Data Science
• Computer Science

Background:

• Imbalanced classification is a significant challenge in data analysis.
• Ensemble techniques are widely used to address data imbalance.
• Recent literature shows diverse approaches to imbalanced classification.

Purpose of the Study:

• To review state-of-the-art ensemble classification algorithms for imbalanced datasets.
• To analyze the incorporation of dynamic selection of base classifiers.
• To introduce a novel patch-ensemble classification method.

Main Methods:

• Evaluation of 14 existing ensemble algorithms with dynamic selection.
• Testing on 56 diverse imbalanced datasets (binary and multi-class).
• Development of a patch-ensemble method combining patch learning and dynamic selection.

Main Results:

• Classical ensemble algorithms with dynamic selection significantly improve classification performance on imbalanced data.
• The proposed patch-ensemble method demonstrates potential for multi-class imbalanced classification.
• Dynamic selection strategies offer a practical approach to enhance ensemble classifiers.

Conclusions:

• Dynamic selection is a viable strategy for improving ensemble classification on imbalanced datasets.
• The novel patch-ensemble method shows promise for complex multi-class imbalanced scenarios.
• Further research into ensemble methods for imbalanced learning is warranted.