direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Page Content

Master Thesis: Anomaly detection on ARIMA manifolds


Anomaly detection on ARIMA manifolds


In this thesis, we propose an approach for anomaly detection in time seris data using the parameters of ARIMA/ARMA models. We use sliding windows to cut the given time series into
many sections and fit an ARIMA/ARMA model on each of these sections (windows). The parameters of the fitted ARMA/ARIMA models serve as features. This is how the dimensionality
reduction is achieved. We then detect anomalies in these features (model parameters) to find out which sections of the original time series data contain anomalies. To detect anomalies in
our feature sets, we use five different unsupervised methods: Robust covariance, ONE-Class SVM, Isolation Forest, Local Outlier Factor and Autoencoders. We show that the proposed
method of performing the anomaly detection in the feature space (with the ARIMA/ARMA model parameters of each window as features) returns promising results, especially when using autoencoders for the anomaly detection.

Supervisor: Boris Lorbeer

Type:  Master Thesis

Duration: 6 months

Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

TU Berlin - Service-centric Networking - TEL 19
Ernst-Reuter-Platz 7
10587 Berlin, Germany
Phone: +49 30 8353 58811
Fax: +49 30 8353 58409