Anomalous Site Detection among Travel Paths as an Unsupervised Learning Problem
Date of Award
Doctor of Philosophy
Industrial and Entrepreneurial Engineering and Engineering Management
Dr. Steven Butt
Dr. Tycho Fredericks
Dr. Lee Wells
Dr. Osama Abudayyeh
Anomalous detection, random paths, unsupervised learning, DBSCAN, spatial scanning
The continuous development of technology has consequently created a highly monitored, tracked, and analyzed environment in which we now live. As a result, a plethora of applications to obtain varying forms of data exist to collect and measure every aspect of our work, operations, possessions, and general lives. A particular area benefiting from the advancement of technological monitoring is the analysis of entity transportation and/or movement patterns. This dissertation investigated relationships among networks of traveling entities, specifically analyzing how points traverse along a random path to define and distinguish between regions of expected or normal activity and locations of anomalous interactions.
Despite the multitude of methods that have been previously discovered to describe and anticipate movement of various entity types (animals, vehicles, ships, tourists, etc.), none have established a distinction between normal movement and anomalous site visits among data lacking historical reference (i.e. normal or anomalous). The actions of this work have developed a labeling methodology to partition and mark individual sites along a random path, with the absence of historical labels. The methodology was developed through an adoption and adaptation of density-based clustering algorithms, spatial scanning boundary constructions, and disjoint network vertex deletions to identify locations of expected behavior and locations of anomalous visits.
This work was motivated by the activities of traveling point locators of Center of Pressure (CP) exerted upon a seated surface, which form a connected path of incremental points. Previous studies of CP have identified incremental CP movements as either large or small, but the line distinguishing the classes has yet to be drawn. The results of this study culminated in a new methodology to detect anomalous sites among unlabeled travel paths. The findings are presented in the context of real-world data and applications.
Restricted to Campus until
Hammond, Megan A., "Anomalous Site Detection among Travel Paths as an Unsupervised Learning Problem" (2019). Dissertations. 3516.