Abstract: To overcome the premature search convergence and the stagnation situation, in this paper, a novel algorithm based on hybrid self-adaptive gravitational search algorithm HSA-GSA is proposed. It can not only maximize the classification accuracy but also select the best subset of features from the data samples. The proposed algorithm combines two solution update strategies to search the solutions, and introduces population reduction method, which effectively balances the exploration search and exploitation search. Moreover, self-adaptive control parameters are designed to reduce the influence on the algorithm due to manually setting. The experimental results on seven practical datasets show that the proposed algorithm HSA has better comprehensive performance with respect to classification accuracy, size of feature subsets and efficiency than original GSA algorithm and similar algorithms. It also proves that the effectiveness of the proposed algorithm.

Key words: gravitational search algorithm, feature selection, self-adaptive, classification method, hybrid optimization

摘要： 特征选择在许多领域具有重要作用，提出一种基于混合自适应引力搜索算法的特征选择方法，在最大化分类精度的同时从数据样本中选出最小特征子集。算法设计两种解更新策略进行组合式搜索，引入群体约简方法，有效地平衡算法的全局搜索和局部收敛能力，同时提出自适应调控参数，减少参数设置对算法性能的影响。在七组真实数据集中的实验结果表明，从分类精度、特征子集大小和运行时间三方面比较，提出的方法优于原始算法和已有相近算法，具有良好的综合性能，是一种有效的特征选择方法。

关键词: 引力搜索, 特征选择, 自适应, 分类算法, 混合优化