Accelerating Outlier Detection with Uncertain Data Using Graphics Processors
Outlier detection (also known as anomaly detection) is a common data mining task in which data points that lie outside expected patterns in a given dataset are identified. This is useful in areas such as fault detection, intrusion detection and in pre-processing before further analysis. There are many approaches already in use for outlier detection, typically adapting other existing data mining techniques such as cluster analysis, neural networks and classification methods such as Support Vector Machines. However, in many cases data from sources such as sensor networks can be better represented with an uncertain model. Detecting outliers with uncertain data involves far more computation as each data object is usually represented by a number of probability density functions (pdfs).
In this paper, we demonstrate an implementation of outlier detection with uncertain objects based on an existing density sampling method that we have parallelized using the cross-platform OpenCL framework. While the density sampling method is a well understood and relatively straightforward outlier detection technique, its application to uncertain data results in a much higher computational workload. Our optimized implementation uses an inexpensive GPU (Graphics Processing Unit) to greatly reduce the running time. This improvement in performance may be leveraged when attempting to detect outliers with uncertain data in time sensitive situations such as when responding to sensor failure or network intrusion.
KeywordsGraphic Processing Unit Outlier Detection Anomaly Detection Intrusion Detection System Uncertain Data
Unable to display preview. Download preview PDF.
- 1.Aggarwal, C.C. (ed.): Managing and Mining Uncertain Data. Springer (2009)Google Scholar
- 2.Alshawabkeh, M., Jang, B., Kaeli, D.: Accelerating the local outlier factor algorithm on a gpu for intrusion detection systems. In: Proceedings of the 3rd Workshop on General-Purpose Computation on Graphics Processing Units (2010)Google Scholar
- 3.Ester, M., Kriegel, H.P., Sander, J., Xu, X.: A density-based algorithm for discoverying clusters in large spatial databases with noise. In: Proceedings of the 2nd International Conference on Knowledge Discovery and Data Mining (1996)Google Scholar
- 6.Bastke, S., Deml, M., Schmidt, S.: Combining statistical network data, probabilistic neural networks and the computational power of gpus for anomaly detection in computer networks. In: 1st Workshop on Intelligent Security (Security and Artificial Intelligence) (2009)Google Scholar
- 7.Huhle, B., Schairer, T., Jenke, P., Strasser, W.: Robust non-local denoising of colored depth data. In: IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Workshop on Time of Flight Camera based Computer Vision (2008)Google Scholar
- 8.Aggarwal, C.C., Yu, P.S.: Outlier detection with uncertain data. In: Proceedings of the SIAM International Conference on Data Mining 2008 (2008)Google Scholar
- 9.Acklam, P.J.: An algorithm for computing the inverse normal cumulative distribution function. Technical report (2003)Google Scholar