简体繁体 English

如何从K最近邻回归中获得结果

[英]How to get results from K Nearest Neighbour Regression

原文 2015-01-14 18:45:24 3 1 canvas/ data-mining/ orange

I want to know the best optimasation for 5 attributes who determine a regressoinal numerical class. 我想知道确定回归数字类别的5个属性的最佳优化。 The K Nearest Neighbour Regression is the best regression method, but how can i get solid results(as in numerical results for the best values for the attributes) based on this regression method? K最近邻回归是最好的回归方法，但是如何基于此回归方法获得可靠的结果（如数值结果中属性的最佳值）？

1 个解决方案

There are several ways to do this depending upon what type of software you use. 有几种方法可以执行此操作，具体取决于您使用的软件类型。 You might want to determine first if indeed K-NN regression is the best algorithm for your needs, assuming you're not overfitting, etc. Some type of Principal Component Analysis (PCA) or other analysis might be in order first to give a sense of attribute priority. 您可能想首先确定K-NN回归是否真的是满足您需求的最佳算法，并假设您没有过拟合等。为了首先给出某种意义，可能需要某种类型的主成分分析（PCA）或其他分析属性优先级。

In Weka, you can do this by embedding (wrapping) the LinearNNSearch method inside an IBk learner. 在Weka中，您可以通过将LinearNNSearch方法嵌入（包装）到IBk学习器中来做到这一点。 In effect, you allow the algorithm to optimize the number of neighbors by means of a repeated, hold-one-out cross-validation process. 实际上，您允许算法通过重复的，保留一出的交叉验证过程来优化邻居数。 Instructions on how to do this is located: 有关如何执行此操作的说明，位于：

http://infochim.u-strasbg.fr/CS3/program/Tutorials/Tutorial2a.pdf http://infochim.u-strasbg.fr/CS3/program/Tutorials/Tutorial2a.pdf