Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
6873265 | Future Generation Computer Systems | 2018 | 11 Pages |
Abstract
The various high-dimensional sensor data can be collected by wireless sensor networks, video monitoring systems and multimedia sensor networks, while High-dimensional sensor data is inherently large-scale because each sensor node has spatial attributes and may also be associated with large amounts of measurement data evolving over time. Detecting outlier in high-dimensional big sensor data is a challenging task. Most of existing outlier detection methods is based on vector representation. However, high-dimensional sensor data is naturally described by tensor representations. The vector-based methods can lead to destroy original structural information and correlation for high-dimensional sensors data, result in the problem of curse of dimensionality, and some outliers cannot be detected. To solve this problem, support high-order tensor data description (STDD) and kernel support high-order tensor data description (KSTDD) are proposed to detect outliers for tensor data. STDD and KSTDD extend support vector data description from vector space to tensor space. KSTDD maintains the structural information of data, avoids the problem caused by the vectorization of tensor data, and improves the performance of outlier detection. Experiments on four sensor datasets show that the proposed method is superior to the traditional vectorized data analysis method.
Related Topics
Physical Sciences and Engineering
Computer Science
Computational Theory and Mathematics
Authors
Xiaowu Deng, Peng Jiang, Xiaoning Peng, Chunqiao Mi,