THESIS
2019
xvii, 113 pages : illustrations ; 30 cm
Abstract
Anomaly detection is an important problem that has been researched within diverse research areas
and application domains, such as fraud detection for financial transactions and intrusion detection
for cyber-security. Many anomaly detection algorithms, including supervised and unsupervised
algorithms, have been proposed since the 19th century. However, the effectiveness of these
techniques are often hinged by three major challenges: (1) there is often a lack of clear boundary
between normal and abnormal observations; (2) high-quality labeled data for training the estimation
model of anomaly detection is usually time-consuming to obtain; and (3) every anomaly analysis
method is specialized for different properties of observations and therefore fits only to some aspects
of the “who...[
Read more ]
Anomaly detection is an important problem that has been researched within diverse research areas
and application domains, such as fraud detection for financial transactions and intrusion detection
for cyber-security. Many anomaly detection algorithms, including supervised and unsupervised
algorithms, have been proposed since the 19th century. However, the effectiveness of these
techniques are often hinged by three major challenges: (1) there is often a lack of clear boundary
between normal and abnormal observations; (2) high-quality labeled data for training the estimation
model of anomaly detection is usually time-consuming to obtain; and (3) every anomaly analysis
method is specialized for different properties of observations and therefore fits only to some aspects
of the “whole truth”. Recent advances in data visualization have shown great promise towards
understanding the relationships among different anomaly detection algorithms, as well as evaluating
the anomaly detection results via intuitive representations of context information.
In this thesis, we present visual anomaly detection techniques and systems to enhance the
exploration of anomaly detection algorithms, and to provide evidence to support or refute the
anomaly detection conclusions in a variety of application areas. Specifically, we present a visual
analytics system named EnsembleLens that evaluates anomaly detection algorithms based on the
ensemble analysis process; we introduce ECGLens, an interactive system for arrhythmia detection and analysis using large-scale ECG data; and we design CloudDet, a unified visual analytics
system that helps cloud service providers interactively detect and diagnose anomalies in cloud
computing systems. We have conducted various quantitative evaluations, case studies, user studies
and expert interviews to demonstrate the effectiveness and usefulness of our proposed systems and
the visualization designs for anomaly detection.
Post a Comment