Opposition-based sine cosine optimizer utilizing refraction learning and variable neighborhood search for feature...

Area of Science:

• Machine Learning
• Data Mining
• Optimization Algorithms

Background:

• Feature selection (FS) is crucial for handling high-dimensional data in machine learning.
• The Sine Cosine Algorithm (SCA) is a metaheuristic effective for continuous optimization but requires adaptation for binary problems like FS.
• Existing binary versions of SCA may lack optimal performance in complex feature selection scenarios.

Purpose of the Study:

• To develop and evaluate improved binary versions of the Sine Cosine Algorithm (SCA) specifically for the feature selection (FS) problem.
• To enhance the performance of the Binary SCA (BSCA) through cumulative improvements incorporating advanced learning and search strategies.
• To rigorously assess the efficacy of the proposed algorithms on diverse real-world datasets, including a COVID-19 dataset.

Main Methods:

• Proposed three cumulative improved binary Sine Cosine Algorithms: IBSCA1 (with Opposition Based Learning), IBSCA2 (adding Variable Neighborhood Search and Laplace distribution), and IBSCA3 (incorporating Refraction Learning).
• Evaluated algorithm performance on 19 real-world datasets using classification accuracy, number of features selected, and fitness values.
• Compared the best proposed algorithm (IBSCA3) against 28 existing popular algorithms.

Main Results:

• IBSCA3 demonstrated superior performance in classification accuracy and fitness values compared to most existing algorithms.
• IBSCA3 achieved a competitive ranking (15th out of 19) for the number of features selected.
• The proposed IBSCA versions, particularly IBSCA3, showed significant improvements over the basic BSCA.

Conclusions:

• The enhanced IBSCA3 algorithm offers a powerful and effective approach for feature selection in machine learning and data mining.
• The integration of Opposition Based Learning, Variable Neighborhood Search, Laplace distribution, and Refraction Learning significantly boosts algorithm performance.
• IBSCA3 represents a state-of-the-art method for feature selection, providing a valuable tool for data analysis and model building.