Data generation and sharing for robust intrusion detection in IoT systems

Important results from the project

The objective of the project was to contribute to the development of ML-based intrusion detection systems for IoT networks, with a focus on knowledge sharing between actors. These objective are fulfilled as the project has delivered a large data set, which is publicly available, as well as several scientific articles which are published internationally. In addition to this, the project has contributed to teaching at Uppsala Universitet by ensuring supervision of 6 student theses (MSc / BSc).

Expected long term effects

The project´s long-term objective was to contribute to an improved Swedish position in the field of cyber security. Interaction with industry is crucial in this context, and here the project has initiated a collaboration with Scaleout Systems AB and had discussions with Ericsson to create awareness about the work. Furthermore, the project´s publications contribute to putting Sweden on the map regarding advanced technology for ML-based intrusion detection.

Approach and implementation

The project ran along two scientific tracks to develop ML-driven IDS for IoT systems: (1) data generation and (2) private knowledge sharing. Within the data generation track, we leveraged and further developed advanced IoT network simulators where attacks were conducted and observed. For the second track, we studied the generalizability of ML models with a particular focus on IDS for IoT systems, and developed methods for knowledge sharing that showed improved attack detection rates.