基于Voronoi面积的密度估计在人口分类中的应用
Classification of Population Using Voronoi Area Based Density Estimation
摘要:
给定两组训练样本,一般的方法是估计密度函数,并根据估计密度的较高值对测试样本进行分类。估计密度的自然方法应该是直方图趋向于频率曲线。但由于不同点的数据积累量不同,在绘制直方图时不能采用相同的区间长度。在这里,我们遵循类似于绘制直方图的方法,目的是进行密度估计和分类,但是没有任意地进行,我们考虑了散点图的图,并获得了基于Voronoi细胞面积的密度估计。但是,任意Voronoi胞的面积可能很难找到,特别是在二维和更高的维度,也在流形。出于这个原因,我们通过绘制许多独立的均匀变量来估计Voronoi单元的面积,这些变量覆盖了训练样本的支持。请注意,虽然我们考虑的是高维和典型流形中的密度估计和分类问题,但我们的模拟和代数计算主要针对实数线上的密度。我们期望在更高的维度和流形的工作将在类似的线,并计划在不久的将来做。此外,我们预计离散的情况下,可以处理略有不同,但类似的线,也留给未来的工作。
英文摘要:
Given two sets of training samples, general method is to estimate the density function and classify the test sample according to higher values of estimated densities. Natural way to estimate the density should be histogram tending to frequency curve. But nature of different amount of data accumulation at different points, obstruct us to take common interval length in drawing histogram. Here we followed a method similar to drawing histogram with an intention to density estimation and classification, but without doing arbitrarily we considered diagram of the scattered plot and has obtained estimate of density based on areas of the Voronoi cells. But area of arbitrary Voronoi cell may be difficult to find out particularly in two and higher dimensions, and also in manifolds. For this reason we estimate the area of the Voronoi cell by drawing many independent uniform variates covering the support of the training samples. Please note that though we have in mind density estimation and classification problem in higher dimension and in typical manifolds, but our simulations and algebraic calculations have been mostly shown for densities on real line. We expect in higher dimension and in manifolds the work will be in similar lines and plan to do in near future. Also we expect discrete cases can be handled in slightly different but similar lines which are also left for future works.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。