Covariance based outlier detection with feature selection.
Covariance based outlier detection with feature selection.
复制标题
基于协方差的异常值检测和特征选择。
DOI：
10.1109/embc.2016.7591264
复制
发表时间：
2016
期刊：
影响因子：
--
通讯作者：
Wang,MichelleY
中科院分区：
文献类型：
--
作者：
Zwilling,ChrisE;Wang,MichelleY
本发明的基于协方差的离群点检测算法从候选特征向量集中选择最擅长识别离群点的特征向量。从生物医学和健康信息学数据中提取的特征可以在疾病评估中提供更多的信息，并且对可以测试的特征的性质和数量没有限制。但是，对于在一组特征上操作的算法来说，一个重要的挑战是从无效特征中筛选出有效的特征。本文描述的强大算法利用时间序列数据中的协方差信息来识别对孤立点识别具有最高灵敏度的特征。实证结果证明了该方法的有效性。
The present covariance based outlier detection algorithm selects from a candidate set of feature vectors that are best at identifying outliers. Features extracted from biomedical and health informatics data can be more informative in disease assessment and there are no restrictions on the nature and number of features that can be tested. But an important challenge for an algorithm operating on a set of features is for it to winnow the effective features from the ineffective ones. The powerful algorithm described in this paper leverages covariance information from the time series data to identify features with the highest sensitivity for outlier identification. Empirical results demonstrate the efficacy of the method.