Description
Smart buildings are considered the future of living. However, at the same time, smart buildings are
threatened by several sophisticated attack vectors. Anomaly detection methods can be utilized as Intrusion
Detection Systems (IDS) to face these threats. With the rapid development of new Machine Learning (ML)
algorithms, they are utilized to solve classification problems in the IDS.
However, such systems often learn from sensitive user data. Privacy-Enhancing Technologies (PETs)
are developed to protect users’ privacy, although they decrease detection capability. This thesis uses
Local Differential Privacy (LDP) on sensitive, but productive features in terms of anomaly detection.
Other application-layer features are dropped to decrease the complexity of the dataset and enable real-time
detection performance.
Using the Edge-IIoT dataset, the approach is tested and evaluated. For this purpose, two other state-ofthe-
art IDS are implemented as well. The proposed model classifies anomalies with an accuracy of 98.17%
while protecting privacy.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching