Binary arithmetic optimization algorithm for feature selection.

Area of Science:

• Computer Science
• Artificial Intelligence
• Optimization

Background:

• Feature selection is crucial for data preprocessing but presents a challenging combinatorial optimization problem.
• Existing meta-heuristic algorithms show promise, but the Arithmetic Optimization Algorithm is limited to continuous problems.

Purpose of the Study:

• To propose novel Binary Arithmetic Optimization Algorithms (BAOAs) for effective feature selection.
• To enhance search speed and improve the ability to escape local optima in feature selection.

Main Methods:

• Developed six BAOAs using different transfer functions to convert continuous search spaces to discrete ones.
• Integrated transfer functions with Lévy flight in six additional algorithms to boost search efficiency and escape local optima.
• Evaluated algorithm performance on 20 University of California Irvine (UCI) datasets.

Main Results:

• BAOA_S1LF emerged as the most superior algorithm among the proposed methods for feature selection.
• Comparative analysis on 26 UCI datasets confirmed the superiority of BAOA_S1LF over other meta-heuristic algorithms.
• Source codes for BAOA_S1LF are publicly available for reproducibility and further research.

Conclusions:

• The proposed BAOAs, particularly BAOA_S1LF, offer a powerful and effective approach to combinatorial optimization in feature selection.
• BAOA_S1LF demonstrates significant advantages in performance and efficiency compared to existing meta-heuristic algorithms.
• The public availability of the source code facilitates broader adoption and advancement in the field of feature selection.