Hyperellipsoidal SVM-Based Outlier Detection Technique for Geosensor Networks

Authors:
Yang Zhang;Nirvana Meratnia;Paul Havinga
Affiliations:
Pervasive Systems Group, University of Twente, Enschede, The Netherlands 7522NB;Pervasive Systems Group, University of Twente, Enschede, The Netherlands 7522NB;Pervasive Systems Group, University of Twente, Enschede, The Netherlands 7522NB
Venue:
GSN '09 Proceedings of the 3rd International Conference on GeoSensor Networks
Year:
2009

Citing 3
Cited 1

Support Vector Data Description

Machine Learning
Estimating the Support of a High-Dimensional Distribution

Neural Computation
Structured One-Class Classification

IEEE Transactions on Systems, Man, and Cybernetics, Part B: Cybernetics

An adaptive and efficient dimension reduction model for multivariate wireless sensor networks applications

Applied Soft Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

Recently, wireless sensor networks providing fine-grained spatiotemporal observations have become one of the major monitoring platforms for geo-applications. Along side data acquisition, outlier detection is essential in geosensor networks to ensure data quality, secure monitoring and reliable detection of interesting and critical events. A key challenge for outlier detection in these geosensor networks is accurate identification of outliers in a distributed and online manner while maintaining low resource consumption. In this paper, we propose an online outlier detection technique based on one-class hyperellipsoidal SVM and take advantage of spatial and temporal correlations that exist between sensor data to cooperatively identify outliers. Experiments with both synthetic and real data show that our online outlier detection technique achieves better detection accuracy compared to the existing SVM-based outlier detection techniques designed for sensor networks. We also show that understanding data distribution and correlations among sensor data is essential to select the most suitable outlier detection technique.