SAX结合Adaboost算法的时间序列分类问题

Research on Time Series Data Classification Combine SAX and AdaBoost Algorithm

SAX是一种典型的符号化特征表示方法.该方法在时间序列特征表示中不仅可以有效地降维、降噪,而且具有简单、直观等特点.时间序列长度不一、特征表示过程中信息损失等问题的存在,使得常规的分类算法难以很好地完成分类任务.在对时间序列数据进行基于SAX符号化的BOP表示方法的基础上,提出了结合集成学习中AdaBoost算法进行分类的新方法,实验结果表明,该方法不仅能很好地处理SAX符号化表示中的信息损失问题,而且与已有方法相比,在分类准确度方面也有了显著的提高.

Symbolic Aggregate approXimation （SAX） is a typical symbolic representation method, which is straight-for- ward and very simple, and it efficiently converts time series data to a symbolic representation with dimension reduction. The is- sues of time series data such as variable in length, and information lose during the representation, making many traditional clas- sification methods unable to apply directly. This paper focus on the SAX discretization method coupled with the Bag of Patterns （BOP） representation in classification task, and proposed the new approach by use AdaBoost Algorithm to remedy the informa- tion loss by SAX representation. The experimental results show that, the approach improved the classification accuracy obvi- ously.