Learning from Imbalanced Data: Integration of Advanced Resampling Techniques and Machine Learning Models for Enhanced...

Area of Science:

• Machine Learning
• Bioinformatics
• Computational Biology

Background:

• Class imbalance is a significant challenge in cancer datasets.
• Accurate classification is crucial for cancer diagnosis and prognosis.

Purpose of the Study:

• Evaluate classification algorithms and resampling methods on cancer datasets.
• Address class imbalance issues in diagnostic and prognostic data.

Main Methods:

• Analyzed five cancer datasets (3 diagnostic, 2 prognostic).
• Employed 19 resampling methods across three categories.
• Utilized 10 classifiers from four distinct categories for comparison.

Main Results:

• Hybrid sampling methods, particularly SMOTEENN, achieved the highest mean performance (98.19%).
• Random Forest (94.69%) was the top-performing classifier, followed by Balanced Random Forest and XGBoost.
• Resampling techniques significantly improved model outcomes compared to the baseline (91.33% performance).

Conclusions:

• Resampling methods are vital for enhancing classification on imbalanced cancer datasets.
• Findings provide insights for integrating machine learning in cancer care.
• Recommends further research into hybrid models and clinical applications for improved cancer diagnosis and prognosis.