Zhu Chenyi, Wang Zhiyi, Peng Yinan, Xiao Wenjun
School of Mechanical and Automotive Engineering, Jinken College of Technology, China.
College of Civil Aviation, Nanjing University of Aeronautics and Astronautics, China.
PLoS One. 2025 May 22;20(5):e0324866. doi: 10.1371/journal.pone.0324866. eCollection 2025.
Feature selection is a crucial preprocessing step in the fields of machine learning, data mining and pattern recognition. In medical data analysis, the large number and complexity of features are often accompanied by redundant or irrelevant features, which not only increase the computational burden, but also may lead to model overfitting, which in turn affects its generalization ability. To address this problem, this paper proposes an improved red-billed blue magpie algorithm (IRBMO), which is specifically optimized for the feature selection task, and significantly improves the performance and efficiency of the algorithm on medical data by introducing multiple innovative behavioral strategies. The core mechanisms of IRBMO include: elite search behavior, which improves global optimization by guiding the search to expand in more promising directions; collaborative hunting behavior, which quickly identifies key features and promotes collaborative optimization among feature subsets; and memory storage behavior, which leverages historically valid information to improve search efficiency and accuracy. To adapt to the feature selection problem, we convert the continuous optimization algorithm to binary form via transfer function, which further enhances the applicability of the algorithm. In order to comprehensively verify the performance of IRBMO, this paper designs a series of experiments to compare it with nine mainstream binary optimization algorithms. The experiments are based on 12 medical datasets, and the results show that IRBMO achieves optimal overall performance in key metrics such as fitness value, classification accuracy and specificity. In addition, compared with nine existing feature selection methods, IRBMO demonstrates significant advantages in terms of fitness value. To further enhance the performance, this paper also constructs the V2IRBMO variant by combining the S-shaped and V-shaped transfer functions, which further enhances the robustness and generalization ability of the algorithm. Experiments demonstrate that IRBMO exhibits high efficiency, generality and excellent generalization ability in feature selection tasks. In addition, used in conjunction with the KNN classifier, IRBMO significantly improves the classification accuracy, with an average accuracy improvement of 43.89% on 12 medical datasets compared to the original Red-billed Blue Magpie algorithm. These results demonstrate the potential and wide applicability of IRBMO in feature selection for medical data.
特征选择是机器学习、数据挖掘和模式识别领域中至关重要的预处理步骤。在医学数据分析中，特征的数量众多且复杂，常常伴随着冗余或不相关的特征，这不仅增加了计算负担，还可能导致模型过拟合，进而影响其泛化能力。为了解决这个问题，本文提出了一种改进的红嘴蓝鹊算法（IRBMO），该算法针对特征选择任务进行了专门优化，并通过引入多种创新的行为策略显著提高了算法在医学数据上的性能和效率。IRBMO的核心机制包括：精英搜索行为，通过引导搜索朝着更有希望的方向扩展来提高全局优化能力；协作狩猎行为，快速识别关键特征并促进特征子集之间的协作优化；以及记忆存储行为，利用历史有效信息提高搜索效率和准确性。为了适应特征选择问题，我们通过传递函数将连续优化算法转换为二进制形式，进一步增强了算法的适用性。为了全面验证IRBMO的性能，本文设计了一系列实验，将其与九种主流二进制优化算法进行比较。实验基于12个医学数据集，结果表明IRBMO在适应度值、分类准确率和特异性等关键指标上实现了最优的整体性能。此外，与九种现有的特征选择方法相比，IRBMO在适应度值方面表现出显著优势。为了进一步提高性能，本文还通过结合S形和V形传递函数构建了V2IRBMO变体，进一步增强了算法的鲁棒性和泛化能力。实验表明，IRBMO在特征选择任务中表现出高效性、通用性和出色的泛化能力。此外，与KNN分类器结合使用时，IRBMO显著提高了分类准确率，在12个医学数据集上与原始红嘴蓝鹊算法相比，平均准确率提高了43.89%。这些结果证明了IRBMO在医学数据特征选择中的潜力和广泛适用性。
