ORCID

0000-0002-3714-2729

Keywords

Time Series Prediction, Time Series Analysis, Anomaly Detection, Text Classification, Transformers, Deep Learning

Abstract

Anomalies are rare in nature. This rarity makes it difficult for models to provide accurate and reliable predictions. Deep learning models typically excel at identifying underlying patterns from abundant data through supervised learning mechanisms but struggle with anomalies due to their limited representation. This results in a significant portion of errors arising from these rare and poorly represented events. Here, we present various methods and frameworks to develop the specialized ability of models to better detect and predict anomalies. Additionally, we improve the interpretability of these models by enhancing their anomaly awareness, leading to stronger performance on real-world datasets that often contain such anomalies. Because our models dynamically adapt to the significance of anomalies, they benefit from increased accuracy and prioritization of rare events in predictions. We demonstrate such capabilities on real-world datasets across multiple domains. Our results show that this framework enhances accuracy and interpretability, improving upon existing methods in anomaly prediction tasks.

Completion Date

2024

Semester

Fall

Committee Chair

Wang, Jun

Degree

Doctor of Philosophy (Ph.D.)

College

College of Engineering and Computer Science

Department

Department of Electrical and Computer Engineering

Degree Program

Computer Engineering

Format

PDF

Identifier

DP0028985

Language

English

Release Date

12-15-2024

Access Status

Dissertation

Campus Location

Orlando (Main) Campus

Accessibility Status

PDF accessibility verified using Adobe Acrobat Pro Accessibility Checker

Share

COinS