一种基于AF的决策树算法

Study on decision tree algorithm based on autocorrelation function

属性序列不同于一般的线性回归模型,其样本点之间存在着一定的相依结构,使得常用的探测异常值的方法,如数据删除、单点求导等,对时间序列而言效果不佳。为了探测时间序列中的强影响点,介绍了同时对几个点作微小扰动时自相关函数(AF)的扰动理论。从应用角度提出一种新的决策树方法,实验结果进一步证实,自相关决策树具有全面性与精确性,从而为进一步实现智能信息检索提供了一种个性化的高效信息检索工具。

Attribute sequence is different from common linear regression model. The common method of detecting abnormal value such as data delete, single point derivation does not have good effectiveness for attribute sequence. In order to detect the strong influence points in attribute sequence, this paper introduces the perturbation theory of autocorrelation function in the mean time having several minute perturbations, On the basis of these, the paper puts forward new decision tree. The experiment and simulation show the effectiveness and accuracy of the decision tree. And the paper presents an improved implement of intelligent information retrieval for people.