基于ISE 的核密度估计和随机置换的单一或协同特征的选择方法

针对数据的特征存在单一和协同特征的选择问题,基于平方误差标准核密度估计和随机置换理论,首先提出一种针对单一特征的特征选择方法(FSKDE-RP);然后,针对协同特征的情况,通过拓展随机置换理论,提出多维协同特征选择算法(SFSKDE-MRP),并利用核神经网络(KNN)分类器的分类精度选择最优特征子集.在模拟数据和真实数据集上的实验结果表明了所提出算法的有效性.

Feature selection method based on the ISE kernel density estimation and random permutation for single or synthetic features

For the single or synthetic features selection problems, based on the integrated square error(ISE) criterion and random permutation, a supervised feature ranking criterion of a single feature is proposed firstly. Then, such a random permutation is extended to synergetic features, and accordingly the synergetic feature selection method is developed. Finally, the optimal feature subset is determined by the classification accuracy obtained by the kernel neural network(KNN) method. Experimental results on synthetic and real datasets show the effectiveness of the proposed algorithm.