用于特征选择的改进二进制蝙蝠算法

Improved binary bat algorithm for feature selection

二进制蝙蝠算法(BBA)是模仿蝙蝠狩猎行为的一种启发式特征选择算法,具有收敛快、模型简单、鲁棒性好的特点。但算法容易出现停滞问题,易陷入局部最优。为此,文中提出了改进的二进制蝙蝠算法(ABBA)。利用种群熵进行传递函数的改进,使得传递函数在适应算法收敛过程的同时赋予算法跳出停滞的能力。其次,加入辅助改进,用于保持算法的收敛性,加速收敛。最后,采用K近邻分类器在22个UCI数据集上进行了与6个较新特征选择算法的对比实验,各项实验结果表明,ABBA的分类准确率和可靠性相比,BA算法均有明显提高,并且在大部分数据集上优于6个其他特征选择算法。ABBA是一种有竞争力的特征选择算法,可以作为机器学习、数据挖掘等领域的有效数据预处理手段。

Binary bat algorithm(BBA)is a heuristic feature selection algorithm that imitates bat hunting behavior,with the characteristics of fast convergence,simple model and good robustness.However,the algorithm is prone to stagnation and easy to fall into local optimum.To this end,this paper proposes an improved binary bat algorithm(ABBA).The transfer function is improved by using population entropy,so that the transfer function adapts to the convergence process of the algorithm while giving the algorithm the ability to jump out of stagnation.Second,auxiliary improvements are added to maintain the convergence of the algorithm and accelerate convergence.Finally,the K-nearest neighbor classifier was used to compare with six newer feature selection algorithms on 22 UCI datasets,and the experimental results showed that the classification accuracy and reliability of ABBA were significantly improved compared with the BA algorithm,and were better than six other feature selection algorithms in most data sets.ABBA is a competitive feature selection algorithm that can be used as an effective means of data preprocessing in areas such as machine learning,data mining,and more.