论文部分内容阅读
针对传统的特征提取方法对于不同分布特征的数据集获得的特征提取效果不佳的问题,提出了结合深度函数的特征提取新方法。通过对传统欧式距离特征提取方法的研究,算法利用核化空间深度函数的思想根据数据集中各个数据特征之间的关系来衡量距离,能更有效地把握数据间距的特征,提取出相似的特征来判断出同类和异类。最后,采用对6个标准UCI数据集进行了3种不同维度和3种不同运行次数的仿真实验,对提出的算法进行了充分验证,实验结果表现该方法有良好的适应性。因此,核化空间深度间距的特征提取方法可以获得较好的特征提取效果,为Relief算法的研究提出了新的思路。
Aiming at the problem that the traditional feature extraction method is not good for the feature extraction of datasets with different distribution features, a new feature extraction method based on depth function is proposed. Through the research on the traditional Euclidean distance feature extraction method, the algorithm uses the idea of nucleation space depth function to measure the distance according to the relationship between each data feature in the data set, and can more effectively grasp the characteristics of data spacing and extract similar features Judge the same type and heterogeneous. Finally, the simulation experiments of three standard UCI datasets are carried out in three different dimensions and three different runs, and the proposed algorithm is fully verified. The experimental results show that the method has good adaptability. Therefore, the feature extraction method of the depth space of the nucleation space can obtain a better feature extraction effect, and brings forward a new idea for the study of the Relief algorithm.