Three Strategies Enhance the Bionic Coati Optimization Algorithm for Global Optimization and Feature Selection...

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Industrial digitization relies on large datasets for model training, but redundant features increase computational costs and reduce generalization.
• Existing optimization algorithms struggle with feature selection (FS) challenges like inadequate global search and suboptimal solutions.

Purpose of the Study:

• To propose the bionic ABCCOA algorithm for enhanced redundant feature elimination in datasets.
• To improve the global search performance and classification accuracy of the Coati Optimization Algorithm (COA) for FS problems.

Main Methods:

• Introduced an adaptive search strategy combining individual learning and disparity learnability to enhance global exploration.
• Developed a balancing factor with phase control and dynamic adjustability to balance exploration and exploitation, avoiding suboptimal subsets.
• Implemented a centroid guidance strategy with population centroid guidance and fractional-order historical memory to improve local exploitation and reduce classification error.

Main Results:

• The bionic ABCCOA algorithm demonstrated an over 90% optimization success rate and faster convergence on test functions and engineering problems.
• Outperformed comparative algorithms across 27 FS problems in key metrics: fitness values, classification accuracy, subset size, and running time.

Conclusions:

• The bionic ABCCOA algorithm is an efficient and robust solution for feature selection, significantly improving dataset preprocessing for industrial digitization.
• The enhancements address critical limitations of the original COA, offering superior performance in complex FS tasks.