Adaptive mechanism-based grey wolf optimizer for feature selection in high-dimensional classification

Area of Science:

• Machine Learning
• Data Mining
• Swarm Intelligence

Background:

• Feature Selection (FS) is vital for enhancing classifier performance by removing irrelevant data.
• Grey Wolf Optimizer (GWO) is a meta-heuristic algorithm effective for optimization but limited in high-dimensional FS.
• GWO can get trapped in local optima, hindering its global search capability.

Purpose of the Study:

• Introduce an Adaptive Mechanism-based Grey Wolf Optimizer (AMGWO) for effective FS in high-dimensional classification.
• Address the limitations of GWO, specifically its susceptibility to local optima and reduced global search capability.
• Enhance the performance of FS algorithms in complex, high-dimensional datasets.

Main Methods:

• Developed AMGWO incorporating a nonlinear parameter control strategy for balanced exploration and exploitation.
• Implemented an adaptive fitness distance balancing mechanism to improve solution selection and search efficiency.
• Integrated an adaptive neighborhood mutation mechanism to dynamically adjust mutation intensity for optimal global search.

Main Results:

• AMGWO demonstrated superior performance across 15 high-dimensional datasets compared to original GWO and its variants.
• Evaluations focused on classification accuracy, feature subset size, and execution speed, highlighting AMGWO's effectiveness.
• The proposed adaptive mechanisms successfully prevented premature convergence and improved the algorithm's ability to find global optima.

Conclusions:

• AMGWO offers a significant advancement in feature selection for high-dimensional classification tasks.
• The adaptive strategies enhance GWO's robustness against local optima and improve overall search efficiency.
• AMGWO provides a superior alternative for optimizing feature selection in machine learning and data mining.