Author : Yanan Sun
Publisher :
Page : 322 pages
File Size : 49,21 MB
Release : 2014
Category :
ISBN :
Spatio-Temporal data is inherently large since each spatial node has spatial attributes and may also be associated with large amounts of measurement data captured over time. In such large and multi-dimensional data identifying anomalies can be a challenge due to the massive data size and relationships among spatial objects. Discovering anomalies in spatio-temporal data is relevant in several domains, such as detecting rare disease outbreaks, detecting oil spills, discovering regions with highway traffic congestion. Most existing techniques for discovering anomalies in spatio-temporal data may find the spatial outliers first and then identify the spatio-temporal anomalies from the data of that specific spatial location. Alternatively, some approaches may discover anomalous time periods and then discover the unusual spatial location in them. This may lead to identifying incorrect spatio-temporal outliers or missing important spatio-temporal phenomena due to the elimination of information after each step. Thus, there is a need to address capturing both space and time simultaneously. A tensor is a multi-dimensional array. It is considered as a powerful tool to manipulate multi-dimensional and multi-variate data. It has a concise mathematical framework for formulating and solving complex data problems efficiently. Tensors can handle complex relationship in spatio-temporal data and their mathematical framework can help us detect spatio-temporal outliers in an effective and efficient manner. Tensors multiplication can integrate spatial and temporal aspects in the data at the same time. In this dissertation, we present our novel approach addressing the key limitation of existing spatio-temporal outlier detection methods by using an efficient tensor based model that supports complex relationships in spatio-temporal data to detect outliers by looking at space and time simultaneously as well as handling the scalability issue when it comes to manipulating large datasets. In this dissertation, we are going to present our novel spatio-temporal tensor model. Based on the spatio-temporal tensor model, we present our clustering-based neighborhood discovery algorithm, neighborhood-based spatial, and spatio-temporal outlier detection algorithms to discover different types of spatio-temporal outliers namely point based and window based outliers. We discuss detailed experimental results for each of the algorithms proposed and also present comparative results.