On the Definition of Intelligible IIoT Architectures

Cunha, Bruno; Sousa, Cristóvão

doi:10.23919/cisti52073.2021.9476342

Cited by 6 publications

(1 citation statement)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…A report entitled "Cyber Signals ISSUE 3" published by Microsoft at the end of 2022 highlighted the increasing trend among organizations to integrate IoT and operational technology (1,2) devices into their operations. While this integration has led to improved production efficiency, it has also exposed vulnerabilities to various forms of attack.…”

Section: Introductionmentioning

confidence: 99%

Feature Selection for Malicious Detection on Industrial IoT Using Machine Learning

Chuang,

Chen

2024

Sensors and Materials

View full text Add to dashboard Cite

The rapid deployment of IoT devices for enhanced convenience and increased production efficiency has resulted in a significant rise in the potential for cyberattacks. Consequently, the detection of malicious attacks has become a crucial concern in industrial IoT (IIoT) applications. Furthermore, IoT usage is continuously expanding, with new functional IoT devices connecting to the network daily, leading to a substantial increase in network traffic. To address the need for an intrusion detection system (IDS) to identify malicious attacks under a high-traffic condition, a highly efficient IDS is essential. In this study, an IDS based on machine learning (ML) with a reduced set of features on the TON_IoT dataset is employed. The TON_IoT includes telemetry data, operating systems data, and network data for an IoT network. A Pearson correlation coefficient (PCC) was applied to assess the correlations among packet features, and a filtering rule based on the Jamovi software's frequency table was used to identify the most essential features within the TON_IoT dataset. Finally, the 45 original features were narrowed down to 10 core features for the IDS to effectively detect intrusion activities. To evaluate the detection performance of malicious intrusion activities using the yielded set of 10 core features, we utilized evaluation metrics including accuracy, precision, recall, and F1-score. Four ML techniques, namely, K-Nearest Neighbors, Random Forest, Naïve Bayes, and eXtreme Gradient Boosting, were tested. The experimental results demonstrated that the four ML techniques could detect multiple types of attack with an accuracy exceeding 96% and with a recall rate over 97%, underscoring the effectiveness and efficiency of utilizing the reduced 10 core features for malicious attack detection while maintaining a high level of accuracy.

show abstract

Section: Introductionmentioning

confidence: 99%