TU Berlin

Service-centric NetworkingYang, Y. (2018). Anomaly Detection with Time Series Analysis. Master Thesis, Technische Universität Berlin


zur Navigation

Es gibt keine deutsche Übersetzung dieser Webseite.

Master Thesis: Anomaly Detection with Time Series Analysis


Anomaly Detection with Time Series Analysis


Log files analyzing has lately received great attention for large computer systems, due to their rich resources for identifying and troubleshooting software failures. Moreover, most log files have voluminous content. This has made it difficult to get the valuable information from large quantities of entries. Also, a single record in log files can not provide enough evidence for failure detection, while the contextual relation of events is neglected. In order to overcome these challenges, a Hidden Markov Model(HMM) anomaly detection approach is developed in this work. In order to take sequential character of data into account, the log file data is clustered into sequences before learning by HMMs. This HMM anomaly detection approach is firstly executed on synthetic data, and then adapted to real-word log data. The conclusion is that, the new developed HMM anomaly detection is able to find out the outliers for both kinds of data, and can be well visualized by t-SNE method.

Supervisor: Tanja Deutsch, Boris Lorbeer

Type:  Master Thesis

Duration: 6 months



Schnellnavigation zur Seite über Nummerneingabe