Internet of Things (IoT) is a promising and thriving technology that incorporates a variety of smart devices that provide enhanced services for remote communication and interaction between humans and physical items. The number of deployed IoT devices will increase to 41.6 billion in 2025, as predicted by International Data Corporation. With such a large population, assaults on IoT networks will harm a vast number of users and IoT devices. In light of this, we explore security from physical and network viewpoints in this thesis.
To preserve privacy in IoT environment, this thesis begins by proposing RASA, a context-sensitive access authorization approach.
We evaluate the promise of RASA-generated policies against a heuristic rule-based policy. The decisions of the RASA and that of the policy are more than 99% consistent.
Furthermore, not only physical attacks but also cybercrimes will threaten IoT networks; consequently, this thesis proposes various Network Intrusion Detection System (NIDS) to identify network intrusions. In this thesis, we firstly examine traditional attacks in the NSL-KDD dataset that can impact sensor networks. Furthermore, in order to detect the introduced attacks, we study eleven machine learning algorithms, among which, XGBoost ranks the first with 97% accuracy.
As attack tactics continue to evolve, Advanced Persistent Threat (APT) poses a greater risk to IoT networks than traditional incursions. This thesis presents SCVIC-APT-2021 to define a APT benchmark. Following upon this, an ML-based Attack Centric Method (ACM) is introduced achieving 9.4% improvement with respect to the baseline performance.
This thesis proposes a Combined Intrusion Detection System (CIDS) that takes network and host information into consideration to reduce data noise and improve the performance of IDS. Two new CIDS datasets, SCVIC-CIDS-2021 and SCVIC-CIDS-2022, are generated. We further propose CIDS-Net to incorporate network and host related data. CIDS-Net boost the macro F1 score of the best baseline by 5.8% (up to 99.95%) and 5.1% (up to 91.3%), respectively on the two datasets.
Besides of detection performance, timely response is considered as a critical metric of NIDS. This thesis introduces Multivariate Time Series (MTS) early detection into NIDS . We form TS-CICIDS2017 which is a time series based NIDS dataset and a new deep learning-based early detection model called Multi-Domain Transformer (MDT) is proposed, resulting in a 84.1% macro F-score with only few of the initial packets.
To reduce the size of NIDS inputs, this work proposes a deep learning-based lossy time series compressor (Deep Dict) to achieve a high compression ratio while limiting the decompression error within a desired range. As demonstrated by the results, Deep Dict outperforms the compression ratio of the state-of-the-art lossy compression methods by up to 53.66%.
| Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/44807 |
| Date | 13 April 2023 |
| Creators | Liu, Jinxin |
| Contributors | Kantarci, Burak |
| Publisher | Université d'Ottawa / University of Ottawa |
| Source Sets | Université d’Ottawa |
| Language | English |
| Detected Language | English |
| Type | Thesis |
| Format | application/pdf |
Page generated in 0.0018 seconds