MSBWO: A Multi-Strategies Improved Beluga Whale Optimization Algorithm for Feature Selection.

Area of Science:

• Machine Learning
• Data Mining
• Optimization Algorithms

Background:

• Feature selection is a critical optimization challenge in machine learning and data mining.
• Metaheuristic methods are increasingly employed to improve feature selection effectiveness.
• Existing methods may struggle with population diversity and escaping local optima.

Purpose of the Study:

• To propose a novel Multi-Strategies Improved Beluga Whale Optimization (MSBWO) algorithm.
• To enhance population diversity and improve the performance of the Beluga Whale Optimization (BWO) algorithm for feature selection.
• To evaluate the effectiveness of MSBWO against conventional and state-of-the-art metaheuristic approaches.

Main Methods:

• Incorporation of Improved Circle Mapping and Dynamic Opposition-Based Learning (ICMDOBL) for population initialization.
• Integration of Elite Pool (EP), Step-Adaptive Lévy Flight and Spiral Updating Position (SLFSUP), and Golden Sine Algorithm (Gold-SA) strategies.
• Comprehensive evaluation on IEEE CEC2005 test functions and ten UCI datasets using a random forest classifier.

Main Results:

• MSBWO demonstrated superior accuracy and a better balance between exploration and exploitation compared to other algorithms.
• The binary MSBWO variant (BMSBWO) achieved competitive classification precision and feature reduction on UCI datasets.
• ICMDOBL, EP, SLFSUP, and Gold-SA strategies collectively enhanced the optimization capabilities.

Conclusions:

• The proposed MSBWO algorithm offers a significant advancement in metaheuristic-based feature selection.
• BMSBWO provides a robust and competitive solution for feature selection in classification tasks.
• The hybrid strategies effectively address limitations of traditional optimization methods in complex feature spaces.