A Comparative Study of Metaheuristic Feature Selection Algorithms for Respiratory Disease Classification

Area of Science:

• Medical Informatics
• Computational Biology
• Artificial Intelligence in Medicine

Background:

• Early diagnosis and treatment of respiratory diseases are crucial for patient health and reducing healthcare costs.
• Automatic respiratory disease detection systems are gaining interest, with machine and deep learning methods being prominent.
• Effective feature selection is vital for the success of machine learning in classifying respiratory diseases.

Purpose of the Study:

• To comparatively analyze six metaheuristic optimization methods for feature selection in respiratory disease classification.
• To evaluate the impact of eight different transfer functions within these metaheuristic methods.
• To assess the performance in both binary and multi-class respiratory disease classification scenarios.

Main Methods:

• Utilized six distinct metaheuristic optimization algorithms.
• Integrated eight different transfer functions with the metaheuristic algorithms.
• Applied these methods to binary and multi-class respiratory disease classification tasks.
• Focused on feature selection to enhance classifier performance.

Main Results:

• Metaheuristic algorithms, when paired with appropriate transfer functions, demonstrated significant data dimensionality reduction.
• The selected feature subsets led to enhanced classification accuracy for respiratory diseases.
• Comparative analysis provided insights into the effectiveness of different metaheuristic and transfer function combinations.

Conclusions:

• Metaheuristic optimization methods are effective tools for feature selection in respiratory disease classification.
• The choice of transfer function significantly influences the performance of metaheuristic algorithms in this domain.
• This research highlights a promising avenue for improving automatic respiratory disease detection systems.