无监督特征选择在时间序列数据挖掘中的应用
Application of unsupervised feature selection in time series data mining
郑宝芬 1苏宏业 1罗林1
作者信息
- 1. 浙江大学智能系统与控制研究所 杭州310027
- 折叠
摘要
疲劳驾驶是交通事故发生的主要原因之一,为了精确预测驾驶员疲劳程度,提出一种基于度量学习的无监督特征选择与识别方法.首先,在对脑电图进行特征提取时,多特征表示的方法克服了传统方法相对单一、往往不能完整表达时间序列信息的缺陷.然后,基于度量学习的特征选择方法对变换之后的特征进行选择,有效降低了样本维度;最后引入支持向量机分类器对其进行分类.通过在公开数据集和真实数据集上对各种过滤式特征选择方法进行的比较实验说明了该方法的有效性.
Abstract
Mental fatigue is a major cause of traffic accidents,in order to predict driver' s fatigue status accurately,an unsupervised feature selection and recognition method based on metric learning is proposed.First,the multi-feature representation method is used in the feature extraction of electroencephalography (EEG),which overcomes the deficiency of traditional methods,such as too simple to completely explain time sequence information.Then,the metric-learning based feature selection method is used to select the features of the transformed data,which reduces the sample dimension efficiently.Finally,the support vector machine(SVM) classifier is used to classify the processed data.Comparison experiments for various filtering feature selection methods on publicly available data sets and real data sets prove the effectiveness of the proposed method.
关键词
脑电图/多特征表示/特征选择/预测Key words
electroencepha lography(EEG)/multi-feature representation/feature selection/forecast引用本文复制引用
出版年
2014
NETL
NSTL
维普
万方数据