常用特征选择方法面临着特征子集空间大小难以确定的问题,取不同的k值,它们的分类效果是相差很大的.粒子群优化算法存在收敛快、获得的是局部最优值而不是全局最优值的问题.针对上述问题,结合云模型的理论知识,提出一种基于云模型的特征选择方法.该算法的适应度函数是通过精确率这一评价指标计算的,将权重分为三个类别来动态确定惯性权重.采用模糊期望交叉熵对原始的特征子集空间进行预选,将预选后的特征子集作为原始特征空间采用改进的特征选择方法,根据模糊期望交叉熵的大小来初始化粒子的种群数及采用迭代变化的阈值作为控制算法的结束条件.实验结果证明了该方法的有效性和可行性.%Common feature selection methods are faced with the problem that the size of feature subset space is difficult to determine. Taking different k values, their classification effects are quite different. PSO algorithm has the problem of fast convergence and obtaining the local optimal value instead of the global optimal value. To solve the above problems, in combination of the theoretical knowledge of cloud models, we propose a feature selection method based on cloud model. The fitness function is calculated by the accuracy rate evaluation index. The weight is divided into three categories to dynamically determine the inertia weight. In this paper, the original feature subset space is preselected using the fuzzy expectation cross entropy, and the pre-selected feature subset is used as the original feature space to adopt an improved feature selection method. According to the size of the fuzzy expected cross entropy, the population of particles is initialized. The iteratively changing threshold serves as an end condition for the control algorithm. The experiment shows that the proposed method is feasible and effective.
展开▼
