Date of Award

Fall 2007

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Electrical & Computer Engineering

Program/Concentration

Electrical Engineering

Committee Director

Min Song

Committee Member

Linda Vahala

Committee Member

Sachin Shetty

Call Number for Print

Special Collections LD4331.E55 G325 2007

Abstract

Wireless sensor networks find several potential applications in a variety of fields, such as environmental monitoring and control, battlefields, surveillance, smart buildings, human health monitoring, etc. These sensor networks consist of a large number of very tiny, inexpensive, and low power sensor nodes, which are deployed in a variety of harsh environments that may result in the sensor data getting corrupted. It is thus critical to detect and report these abnormal values in the sensor data, in order to have a better understanding of the monitored environment. Detection of the abnormal values is of special interest for the sensor network research community because it helps in the identification of certain events of interest, malicious activities, or faulty nodes in the network.

In contrast to a centralized scheme of outlier detection, a distributed approach deals with the detection of outliers at the granularity of the sensor nodes itself. This thesis addresses the problem of outlier detection in a sensor network scenario. The main contribution of this thesis is to develop a distributed cluster based outlier detection algorithm for sensor data. At each node, the local data is first partitioned into a fixed number of clusters based on the principles of the popular k-means clustering paradigm. The cluster summaries are then sent to few higher capacity nodes in the network. It is at these high capacity nodes that the outlier clusters are detected using the K-nearest neighbor algorithm by identifying the K closest clusters. The technical challenge, however, is to be able to achieve the goal of outlier detection in the network, i.e. at the node level itself. This is because the transmission of raw data from all the nodes to a centralized location for further analysis may result in increased energy consumption and communication overhead. In order to address the issue of achieving outlier detection in sensor data with minimized energy and communication costs, a distributed scheme of outlier detection is used. In such a scheme, only a summary of cluster information is being transmitted across the network, rather than the entire set of raw data. This greatly reduces the number as well as the size of messages being transmitted across the network and, thereby, results in reduced communication costs. A set of simulation experiments were performed and results obtained. The performance of the proposed approach is then studied based on varying a set of system parameters.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/gb5b-n639

Download

Share

COinS