A robust binary secretary bird optimization method for high-dimensional data classification.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Feature selection is crucial for machine learning (ML) in medical applications.
• High-dimensional datasets can reduce ML model accuracy and interpretability.
• Existing feature selection methods may not offer optimal performance.

Purpose of the Study:

• Introduce the Binary Secretary Bird Optimization Algorithm (B-SBOA) for effective feature selection.
• Evaluate B-SBOA's performance against established binary metaheuristic algorithms.
• Demonstrate B-SBOA's capability in improving classification accuracy and reducing feature dimensionality.

Main Methods:

• Developed B-SBOA, a binary variant of the secretary bird optimization algorithm.
• Translated secretary bird behaviors into binary search strategies for exploration-exploitation balance.
• Tested B-SBOA on 25 UCI benchmark datasets, comparing it with nine other algorithms (PSO, GWO, MPA, HBO, SMA, SFOA, DOA, SCA, MSO).

Main Results:

• B-SBOA demonstrated superior or competitive performance across F-score, precision, and recall metrics.
• Achieved average improvements of 3-8% in F-score and 2-6% in precision and recall.
• Outperformed existing methods on high-dimensional datasets like Arrhythmia and Hillvalley, improving classification accuracy and reducing feature count.

Conclusions:

• B-SBOA is an effective metaheuristic algorithm for feature selection in machine learning.
• The algorithm offers a balanced approach to exploration and exploitation for optimal feature subset identification.
• B-SBOA shows significant potential for enhancing ML model performance in complex, high-dimensional biomedical datasets.