基于自适应环境因子熵权决策的多目标特征选择

Adaptive environmental factor entropy weight decision-making-based multi-objective feature selection

多目标进化算法在特征选择方面有显著的优势,但其求解高维数据最优特征子集的性能依然较差,且从获得的Pareto解集中选择合理最优解仍是一个挑战性的问题.为了解决该问题,提出一种基于自适应环境因子熵权决策的多目标特征选择算法.首先,通过设计环境因子来自适应识别关键特征,优化候选特征子空间;其次,将环境因子嵌入改进的交叉算子和变异算子,实现全局最优特征子集的自适应搜索;最后,利用关联环境因子的熵权决策策略,从获得的Pareto解集中选出最优解.实验表明,与现有的五种多目标特征选择算法相比,提出的算法具有更高的分类精度,并能准确地获取全局最优解,验证了该算法的有效性.

Multi-objective evolutionary algorithms have significant advantages in feature selection,but their performance in solving the optimal feature subset of high-dimensional data is still poor,and selecting a reasonable optimal solution from the obtained Pareto solution set remains a challenging issue.To solve this problem,this paper proposes a multi-objective feature selection algorithm based on adaptive environmental factor entropy weight decision-making.The advantages of the proposed algorithm are as follows.Firstly,key features are adaptively identified by designing environmental factors to optimize candidate feature subspaces.Secondly,environment factors are embedded into improved crossover and mutation operators to achieve adaptive search for the globally optimal feature subset.Finally,using an entropy weight decision-making strategy that correlates environmental factors,the optimal solution is selected from the obtained Pareto solution set.Experiments show that the proposed algorithm has higher classification accuracy and accurately obtains the global optimal solution compared to the existing five multi-objective feature selection algorithms,verifying the effectiveness of the proposed algorithm.