基于时点分割的核Fisher判别分析–顺序回归机多类分类建模方法

The kernel Fisher discriminant analysis-order regression machine multi-class classification modeling method based on time point partition

中文关键词  多维时间序列  核Fisher判决分析  顺序回归机  指数平滑法

英文关键词  multidimensional time series  KFDA  ORM  exponential smoothing model

中文摘要

      针对多维时间序列的多类分类问题, 本文提出基于时点分割思想的核Fisher判别分析–顺序回归机(KFDA–ORM)多类分类建模方法. 该方法利用核Fisher判别分析(KFDA)与顺序回归机(ORM)的互补性得到分类决策函数; 对分类样本的多维时间序列进行时点分割处理, 使用决策函数得到各时点的分类级别; 通过指数平滑分析得到采样周期内样本的最终分类结果. 通过实例验证, 该方法对多维时间序列的分类具有较好效果, 是一种有效的多类分类方法.

英文摘要

      To tackle the multi-class classification problem of multidimensional time series, we present a kernel Fisher discriminant analysis (KFDA) and order regression machine (ORM) multi-class classification model based on the time point partition. The classification decision function of this method is obtained from the complementarities of KFDA and ORM. The time point segmentation is used to deal with the multidimensional time series of the samples. The classification level at each time point is determined by the decision function. The final classification results of the samples in the sampling period are obtained by using the exponential smoothing. The experimental results show that the algorithm has desirable results in the classification of multi-dimensional time series, and provides an effective multi-class classification model.