Imbalanced Data Classification via Cooperative Interaction Between Classifier and Generator.

Area of Science:

• Machine Learning
• Artificial Intelligence
• Data Science

Background:

• Imbalanced datasets pose significant challenges in machine learning, often leading to biased classifiers favoring the majority class.
• Generative Adversarial Networks (GANs) have been explored to address data imbalance, but existing methods lack optimal integration between the classifier and generator components.

Purpose of the Study:

• To propose a novel three-player framework integrating a discriminator, generator, and classifier for enhanced imbalanced data classification.
• To improve the utilization of the relationship between the classifier and generator for more effective minority class data generation.

Main Methods:

• A novel three-player structure comprising a discriminator, generator, and classifier is introduced.
• Decision boundary regularization is incorporated into the training process.
• The generator is trained collaboratively with the classifier to generate synthetic minority samples that expand the decision region.

Main Results:

• The proposed method effectively addresses the bias in classifiers trained on imbalanced data.
• The cooperative training of the generator and classifier leads to a gradual expansion of the minority decision region.
• Experimental results demonstrate superior performance compared to existing methods on both synthetic and real-world imbalanced datasets.

Conclusions:

• The novel three-player GAN framework offers a significant advancement in handling imbalanced data classification.
• The proposed cooperative training strategy enhances the generation of informative minority samples, improving classifier performance.
• This approach provides a robust solution for real-world imbalanced classification tasks.