Enhancing and improving the performance of imbalanced class data using novel GBO and SSG: A comparative analysis

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Class imbalance poses a significant challenge in machine learning, leading to biased models and poor performance on minority classes.
• Traditional oversampling methods like Synthetic Minority Oversampling Technique (SMOTE) can create overlapping synthetic samples, exacerbating bias.
• Generative Adversarial Networks (GANs) show promise for generating realistic data but are complex to train.

Purpose of the Study:

• To propose novel techniques that effectively address the class imbalance problem.
• To overcome the limitations of existing oversampling methods, particularly SMOTE.
• To enhance the performance of machine learning models on imbalanced datasets.

Main Methods:

• Development of two new techniques: GAN-based Oversampling (GBO) and Support Vector Machine-SMOTE-GAN (SSG).
• Evaluation of GBO and SSG on nine imbalanced benchmark datasets.
• Comparison of proposed methods against existing SMOTE-based approaches.

Main Results:

• SSG and GBO demonstrated superior performance compared to several existing SMOTE-based methods on benchmark datasets.
• The proposed SSG and GBO methods achieved over 90% accuracy in classifying minority classes across varying test data percentages (20%, 30%, 40%).
• SSG generated synthetic minority samples exhibiting Gaussian distributions, a characteristic often challenging to achieve with standard SMOTE or SVM-SMOTE.

Conclusions:

• The novel GBO and SSG techniques effectively mitigate the class imbalance problem.
• These methods offer a significant improvement in minority class classification accuracy.
• SSG's ability to generate Gaussian-distributed synthetic data enhances its utility for imbalanced learning tasks.