Synthetic Boosted Resampling Using Deep Generative Adversarial Networks: A Novel Approach to Improve Cancer...

Area of Science:

• Machine Learning in Healthcare
• Data Science
• Medical Informatics

Background:

• Imbalanced datasets pose significant challenges in critical healthcare applications like cancer diagnosis and prognosis.
• Traditional resampling methods may not adequately address the complexities of generating realistic synthetic data for minority classes.
• Effective handling of imbalanced data is crucial for accurate medical diagnoses and prognoses.

Purpose of the Study:

• To evaluate the effectiveness of Generative Adversarial Networks (GANs) as a resampling method for imbalanced healthcare datasets.
• To compare the performance of various classifier models (Boosting, Bagging, Linear, Non-linear) when using GAN-generated data.
• To assess the impact of GAN-based resampling on key performance metrics, including accuracy, precision, recall, F1 score, and ROC AUC.

Main Methods:

• Generative Adversarial Networks (GANs) were employed to generate synthetic data for the minority class, addressing class imbalance.
• Four distinct classifier types—Boosting, Bagging, Linear, and Non-linear—were evaluated.
• Performance was rigorously assessed using standard classification metrics, with a focus on Receiver Operating Characteristic Area Under the Curve (ROC AUC).

Main Results:

• Baseline classification without resampling demonstrated significant performance limitations.
• GAN-based resampling substantially improved the detection of minority instances and overall classification accuracy.
• Average ROC AUC increased from approximately 0.8276 to over 0.9734, with the GradientBoosting classifier achieving the highest ROC AUC of 0.9890.

Conclusions:

• Generative Adversarial Networks (GANs) represent a powerful strategy for resampling imbalanced healthcare datasets.
• Advanced classifier models, particularly Boosting and Bagging, show superior performance when combined with GAN-based resampling.
• The study confirms the efficacy of GANs in enhancing predictive accuracy for critical healthcare applications.