有序判别分析新算法及其应用

判别分析是用已知分类数据建模对未知分类数据进行判别的方法,所用数据和分类不分顺序。要对有序又有周期数据进行判别分析,就要探索有序判别的新方法。这种方法的分类应当是有序的,并且能够排除事物发展周期性的干扰。本文介绍多元数据有序判别分析新方法的原理、建模流程、应用流程和应用实例。这种判别分析将分类建模与判别归类分开。新方法对多元数据建模时在多类模型中建立滑移的多套子模型,应用时根据应用领域的知识对样本归属作初步预估,然后程序选择相关的子模型进行判别归类。这种方法解决了由于时间序列多元数据周期性造成的样本分类颠倒问题,为时间序列数据的分类和预测开辟了新途径,在实际应用中取得了良好的效果,解决了重大难题。

On New Arithmetic Method of Sequent Discrinant Analysis and its Application

Discriminant analysis is a method which classify the type-unknown data by modeling the type-known data,in which data and types are not ordinal.In order to discriminate ordinal and periodic data,new discriminant analysis method should be explored.This new method should classify data in sequence and eliminate the disturbance of periods.This paper introduces the principle,modeling flow chart,applying flow chart and a practical example of new arithmetic method called Sequent Discriminant Analysis(SDA) which may be used in multivariate sequence data.In SDA,the class modeling and data discriminating are separated.While modeling to multivariate data by this kind of discriminant analysis,a number of child-models are built in the model by way of slippage.While applying the model,the initial estimation of the samples' classification should be given according to the knowledge in the problem-corresponding field.Then the program selects the appropriate child-models to discriminate the classes. In this way,we solved the upside down problem of sample classification caused by the periodicity of multivariate time series data.Thus,a new approach is made to classify and forecast the time series data.In practical application,we have achieved a lot and given important problems approving solutions.