A multi-learning strategy golden eagle optimization algorithm for feature selection in data classification

ZD Li and L Yu and YY Huang and HY Zhao, INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS, 16, 8237-8267 (2025).

DOI: 10.1007/s13042-025-02721-x

Feature selection is crucial for enhancing the performance and efficiency of machine learning models, as well as for developing robust and scalable machine learning applications. It involves identifying a subset of features from a larger set that not only exhibits the lowest internal similarity but also the highest relevance to the target class. Therefore, it is fundamentally an optimization problem. The original golden eagle optimization (GEO) algorithm encounters challenges like slow convergence speed and low solution accuracy in feature selection. This paper proposes a multi-learning strategy GEO (MLSGEO) algorithm to overcome these challenges. The MLSGEO integrates example-pool learning, random learning, and worst-best learning strategies into the prey selection process to enhance the search capability and prevent the algorithm from falling into local optima. Meanwhile, a sine-cosine parameter adjustment strategy is introduced for the cruise and attack coefficients to balance global search and local exploitation capabilities. Furthermore, crossover and mutation are applied to prevent premature convergence. Numerical experiments have been conducted on CEC 2017 test suite and 15 feature selection datasets and compared against several advanced meta-heuristic methods. Experimental results show that MLSGEO displays a better performance with respect to its competitors. Also, the effectiveness of three proposed strategies has been empirically quantified.

Return to Publications page