Date of Award


Document Type


Degree Name

Master of Science


Department of Electrical and Computer Engineering

First Advisor

Richard K. Martin, PhD.


Current research in employing pattern recognition techniques in a wireless sensor network (WSN) to detect anomalous or suspicious behavior is limited. The purpose of this research was to determine the feasibility of an accurate tracking and intent assessment system of unknown or foreign radio frequency (RF) emitters in close proximity to and within military installations as a method for physical security. 22 position tracks were collected using a hand-held Global Positioning System (GPS) unit and a training data set from five different features was generated for each position track. Each collected position track was individually classified as suspicious or non-suspicious by the leave-one-out-cross-validation (LOOCV) method using four different classification methods. The four classification methods used in this research were the linear discriminant function (LDF), the diagonal linear discriminant function (DLDF), the quadratic discriminant function (QDF) and the Mahalanobis distance method. The accuracies and false positive/negative error rates of the four classification methods were compared for different assessment system configurations. Additionally, best fit receiver operating characteristic (ROC) curves were generated for each classification method and discussed. The QDF classification method out-performed the other three classification methods. This classification method achieved an accuracy of 95% when it classified the 22 position tracks one at a time. The lowest false positive and false negative rates were 10% and 0%, respectively. The prior probabilities for the non-suspicious and suspicious classes were both set to 50% class for this configuration.

AFIT Designator


DTIC Accession Number