2022 · ACM/IFIP/USENIX Middleware 2022 Demos/Posters
Deep Learning Driven Anomaly Based Intrusion Detection System for IoT: Poster Abstract
Evidence basis: full-text-reviewed · Review status: catalog-reviewed; paper-author approval pending
iot-security anomaly-detection machine-learning deep-learning-systems
IoTID20 RNN SMOTE PSO binary classification multiclass classification IoT attacks intrusion detection
Core contribution: The poster proposes a hybrid IoT intrusion-detection pipeline with binary anomaly detection followed by multiclass attack classification.
Problem and motivation
The poster motivates anomaly/intrusion detection for growing IoT networks, arguing that existing methods may be poorly tuned and that large feature sets increase computation, memory, and training time (p. 1, "Abstract/Introduction").
Method and contribution
A hybrid pipeline uses a machine-learning binary classifier for normal/anomalous traffic and an RNN for multiclass attack-type classification. It applies SMOTE for imbalance, PSO feature selection, and hyperparameter tuning over loss, optimizer, batch size, and epochs (p. 1, "Methodology").
Findings and evidence
IoTID20 with 86 captured features is named. The conclusion claims that feature selection and balancing such as SMOTE and "SMO" can improve performance and execution time, but the poster reports no numeric accuracy, precision, recall, F-score, latency, split, baseline, hardware, or run count.
Limitations and future directions
Limitations: Classifier names beyond "machine learning binary classifier" and "RNN," software versions, traffic-generation procedure, hardware, train/test protocol, and numeric results are unknown. The methodology names PSO, while the conclusion says SMO; preserve this source inconsistency.
Future work: More realistic datasets/attacks, other ML/DL models, GPU/cloud acceleration, and a high-performance GPU platform (p. 1, "Future Work").
Sources and identifiers
- Published version published
- Public poster artifact · PDF public_full_text
When to cite this paper
Cite this paper when your work uses or compares the poster-level proposal of a two-stage IoT IDS combining binary anomaly screening with RNN attack-type classification.
- For the poster-level proposal of a two-stage IoT IDS combining binary anomaly screening with RNN attack-type classification.
- For the stated use of IoTID20/86 features with SMOTE and PSO feature selection in an IoT intrusion-detection pipeline.
- For a research motivation/future-work citation on reducing IoT IDS feature-processing cost through selection, balancing, and GPU/cloud acceleration; do not cite it for a numeric performance result.
Citation
@inproceedings{ezzatiJivan2022deeplearning,
author = {Yue Guan and Naser Ezzati-Jivan},
title = {Deep Learning Driven Anomaly Based Intrusion Detection System for IoT: Poster Abstract},
year = {2022},
booktitle = {ACM/IFIP/USENIX Middleware 2022 Demos/Posters},
pages = {19-20},
publisher = {ACM},
doi = {10.1145/3565386.3565493},
url = {https://doi.org/10.1145/3565386.3565493}
}Other citation formats for Word and reference managers
Guan, Y., & Ezzati-Jivan, N. (2022). Deep Learning Driven Anomaly Based Intrusion Detection System for IoT: Poster Abstract. In ACM/IFIP/USENIX Middleware 2022 Demos/Posters (pp. 19-20). https://doi.org/10.1145/3565386.3565493Y. Guan and N. Ezzati-Jivan, "Deep Learning Driven Anomaly Based Intrusion Detection System for IoT: Poster Abstract," in ACM/IFIP/USENIX Middleware 2022 Demos/Posters, pp. 19-20, 2022, doi: 10.1145/3565386.3565493