Binary dwarf mongoose optimizer for solving high-dimensional feature selection problems.

Area of Science:

• Machine Learning
• Data Science
• Optimization Algorithms

Background:

• Feature selection is crucial for improving machine learning model accuracy by removing irrelevant or redundant data.
• High-dimensional datasets pose challenges for traditional feature selection methods.
• Advanced optimization techniques are needed to identify optimal feature subsets efficiently.

Purpose of the Study:

• To introduce a novel binary version of the dwarf mongoose optimization algorithm (BDMO) for high-dimensional feature selection.
• To evaluate the effectiveness and superiority of the BDMO algorithm against established feature selection techniques.
• To demonstrate the capability of BDMO in improving classification performance and model accuracy.

Main Methods:

• Development of the binary dwarf mongoose optimization (BDMO) algorithm.
• Validation using 18 high-dimensional datasets from the Arizona State University feature selection repository.
• Comparative analysis of BDMO against other well-known feature selection methods.

Main Results:

• BDMO achieved the best fitness values in 14 out of 18 datasets (77.77%).
• BDMO demonstrated high stability, yielding the lowest standard deviation in 13 out of 18 datasets (72.22%).
• BDMO attained higher validation accuracy in 15 out of 18 datasets (83.33%), outperforming other methods on COIL20 and Leukemia datasets.

Conclusions:

• The BDMO algorithm is a highly effective and stable method for high-dimensional feature selection.
• BDMO significantly enhances classification accuracy compared to existing techniques.
• The proposed BDMO approach represents a superior solution for optimizing feature subsets in machine learning.