Hybrid clustering strategies for effective oversampling and undersampling in multiclass classification

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Real-world datasets often exhibit multiclass imbalance, where rare classes have few samples.
• This imbalance poses significant challenges for predictive modeling and analysis.
• Existing methods struggle to effectively manage diverse levels of class disparity.

Purpose of the Study:

• To introduce a novel hybrid cluster-based oversampling and undersampling (HCBOU) technique.
• To address the challenges posed by multiclass imbalanced datasets.
• To improve classification performance on datasets with rare occurrences.

Main Methods:

• The HCBOU algorithm clusters data, separating majority and minority classes.
• It employs oversampling for minority classes and undersampling for majority classes.
• Classification is performed using one-vs-one and one-vs-all decomposition schemes.

Main Results:

• The proposed HCBOU algorithm demonstrated superior performance compared to state-of-the-art methods.
• Experiments on 30 datasets confirmed the algorithm's effectiveness across various scenarios.
• HCBOU showed robust performance across different levels of class imbalance.

Conclusions:

• The HCBOU algorithm is highly effective in handling multiclass imbalanced datasets.
• It offers a significant improvement over existing techniques for imbalanced data.
• The method provides a robust solution for real-world applications with rare events.