ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume II-4/W2
https://doi.org/10.5194/isprsannals-II-4-W2-67-2015
https://doi.org/10.5194/isprsannals-II-4-W2-67-2015
10 Jul 2015
 | 10 Jul 2015

INCREMENTAL PRINCIPAL COMPONENT ANALYSIS BASED OUTLIER DETECTION METHODS FOR SPATIOTEMPORAL DATA STREAMS

A. Bhushan, M. H. Sharker, and H. A. Karimi

Keywords: Outlier Detection, Incremental Principal Component Analysis, Spatiotemporal Data Streams

Abstract. In this paper, we address outliers in spatiotemporal data streams obtained from sensors placed across geographically distributed locations. Outliers may appear in such sensor data due to various reasons such as instrumental error and environmental change. Real-time detection of these outliers is essential to prevent propagation of errors in subsequent analyses and results. Incremental Principal Component Analysis (IPCA) is one possible approach for detecting outliers in such type of spatiotemporal data streams. IPCA has been widely used in many real-time applications such as credit card fraud detection, pattern recognition, and image analysis. However, the suitability of applying IPCA for outlier detection in spatiotemporal data streams is unknown and needs to be investigated. To fill this research gap, this paper contributes by presenting two new IPCA-based outlier detection methods and performing a comparative analysis with the existing IPCA-based outlier detection methods to assess their suitability for spatiotemporal sensor data streams.