Mitigating IoT Security Vulnerabilities Using Artificial Intelligence Approaches

The system proposed in this paper focuses on providing an improved security solution to mitigate threats in IoT (Internet Of Things) systems. IoT works on connecting digital devices assigned with unique identifiers to exchange data over a network without real-time human-to-human or human-to-computer interaction. However, IoT devices pose severe security risks for their ease of use. Thus, the paper proposes a two-stage-based Intrusion Detection System (IDS) that allows users to protect their IoT devices from vulnerabilities at network levels. The improvement from the existing system on security monitoring is that this prototype system is a lightweight, simpler architecture that can be implemented within in-house hardware and in cloud solutions. For the first stage, Signature-based IDS (SIDS), the system outperformed the previous system with an accuracy of $\mathbf{99.4} \%$ with the XGBOOST algorithm on the UNSW15_NB15 Dataset. It is trained to save known malicious patterns. If the packet header matches any pattern, SIDS classifies it and blocks the attack. The second stage, Anomaly-based IDS (AIDS), uses Artificial Intelligence algorithms, LSTM, and 1D CNN to form a hybrid model. It was tested with two datasets, UNSW15_NB15 Dataset and BotIoT Dataset, which had $100 \%$ and $86.4 \%$ accuracy, respectively. If SIDS gets an unknown pattern, it will be passed to AIDS. This will count the deviation between the incoming packet and the normal packet. If the deviation exceeds the normal threshold, it will flag the packet as a vulnerability. The research also focuses on classifying IoT security threats into two main categories: Network-based and Hardware-based.