Adaptive Swarm Balancing Algorithms for rare-event prediction in imbalanced healthcare data.

Area of Science:

• Computational biology
• Data science
• Medical informatics

Background:

• Clinical data analysis is crucial for disease control but often suffers from imbalanced class distributions, where positive samples are rare.
• Existing methods for rebalancing imbalanced datasets may not scale effectively to large datasets, limiting their practical application in medical contexts.

Purpose of the Study:

• To develop and evaluate novel methods for rebalancing binary imbalanced datasets, particularly those encountered in medical applications.
• To investigate the efficacy of meta-heuristic algorithms, specifically particle swarm optimization and bat algorithm, in enhancing the synthetic minority over-sampling technique (SMOTE).

Main Methods:

• Two approaches were investigated: processing the entire dataset and adaptively processing segmented datasets using meta-heuristic algorithms to optimize SMOTE parameters.
• The adaptive segmentation approach, termed Adaptive Swarm Balancing Algorithms, was compared against processing the full dataset.
• Meta-heuristic algorithms were employed to optimize key parameters within SMOTE.

Main Results:

• Processing the full dataset with meta-heuristic algorithms did not yield scalable performance improvements for larger data scales.
• Adaptive Swarm Balancing Algorithms demonstrated significant improvements in both efficiency and effectiveness on large imbalanced datasets.
• The proposed methods resulted in more credible classifier performances and reduced computational run times compared to brute-force approaches.

Conclusions:

• Adaptive Swarm Balancing Algorithms offer a scalable and effective solution for rebalancing large imbalanced medical datasets.
• Optimizing SMOTE parameters using meta-heuristic algorithms enhances classifier performance and efficiency, making it suitable for real-world medical data challenges.