Enhancing Anomaly Detection in Time Series Data Using Hybrid Deep Learning Methods

Pallikona, Wonder Paul (2024) Enhancing Anomaly Detection in Time Series Data Using Hybrid Deep Learning Methods. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (2MB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Time series anomaly detection is a critical task for detecting irregularities or deviating patterns in the data which may indicate system failure, fraud, etc. This research focuses on detecting anomalies using various deep learning techniques, including LSTM, GRU, RNN based Autoencoders. The New York City taxi data is employed in the study to detect anomaly patterns in time series data. Deep learning models like RNN-Autoencoder and its variants such as LSTM and GRU based autoencoders are trained in unsupervised setting by minimizing the reconstruction error, and an anomaly is found when the reconstruction error is above certain specified threshold. Although the main goal of this research is not to make accurate predictions but predicting anomalies in the dataset. This research evaluates the various auto encoder variants based on MAE, MSE and RMSE based on reconstruction error on test data. The analysis of the results illustrates that the RNN Autoencoder has the lowest MAE, MSE, and RMSE scores of 0.0457, 0.0036, and 0.0600, respectively with the highest R² score of 0.8897. The LSTM Autoencoder also shows high performance, while the GRU Autoencoder, which has the advantage of having lower computational complexity provides relatively lower performance. The results of the proposed hybrid deep learning models are then compared with a more conventional approach, the Isolation Forest, to demonstrate the strengths of deep learning in identifying nonlinear temporal relationships in time series data. The results show that by using deep learning methods, anomaly detection performance can be enhanced by a huge margin that provides affordable solutions for real-time monitoring and decision-making in numerous domains including operational excellence, fraud prevention, and predictive maintenance.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Jilani, Musfira UNSPECIFIED
Uncontrolled Keywords:	Anomaly Detection; LSTM; RNN; GRU-Autoencoders; Isolation Forest
Subjects:	Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science Q Science > QA Mathematics > Computer software > Computer Security T Technology > T Technology (General) > Information Technology > Computer software > Computer Security Q Science > Q Science (General) > Self-organizing systems. Conscious automata > Machine learning
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	04 Sep 2025 08:32
Last Modified:	04 Sep 2025 08:32
URI:	https://norma.ncirl.ie/id/eprint/8765

Actions (login required)

View Item