Enhanced IoT Intrusion Detection through Machine Learning

Strategic Insights -

New research conducted by the University of Jeddah has yielded promising results in the field of Machine Learning, specifically in the development of a smart deep learning model for enhanced IoT intrusion detection. The study aimed to address the limitations of existing approaches in processing large volumes of data and evolving cyber threats. The research team successfully developed and tested a novel system that combines XGBoost and Sequential Neural Network (OSNN) algorithms, augmented with various filters and techniques, to improve multiclass intrusion detection across multiple datasets.

Key Takeaways:

  • The proposed system was tested on three challenging datasets: NSL-KDD, UNSW-NB15, and CICIDS2017, with optimized XGBoost and OSNN models achieving high accuracy and F1-scores on the NSL-KDD dataset.
  • The optimized XGBoost model achieved an accuracy of 99.93%, F1-score of 99.84%, and MCC of 99.86% on the NSL-KDD dataset.
  • The optimized SNN model achieved an accuracy of 99.0% and AUC of 1.00 on the NSL-KDD dataset.
  • The OSNN model performed exceptionally well on the UNSW-NB15 dataset, with an accuracy of 96.80% and a loss of 0.0777.
  • The OSNN model also performed well on the CICIDS-2017 dataset, with an accuracy of 99.53% and a loss of 0.0236.
  • The research highlights the potential of the proposed method to enhance intrusion detection, system integrity, fraud prevention, and overall network performance.
  • Faisal S. Alsubaei, the lead researcher from the University of Jeddah, emphasized the importance of developing robust security measures to counter increasingly sophisticated cyber threats.

Statistics:

  • The optimized XGBoost model achieved an excellent accuracy of 99.93% on the NSL-KDD dataset.
  • The optimized SNN model achieved an accuracy of 99.0% and AUC of 1.00 on the NSL-KDD dataset.
  • The OSNN model achieved an accuracy of 96.80% and a loss of 0.0777 on the UNSW-NB15 dataset.
  • The OSNN model achieved an accuracy of 99.53% and a loss of 0.0236 on the CICIDS-2017 dataset.
  • The proposed system was tested on three diverse datasets, demonstrating its versatility and robustness.

Sources:

  • Smart deep learning model for enhanced IoT intrusion detection. Scientific Reports, 2025;15(1):20577. Scientific Reports can be contacted at: Nature Portfolio, Heidelberger Platz 3, Berlin, 14197, Germany. (Nature Publishing Group - www.nature.com/; Scientific Reports - www.nature.com/srep/)
  • University of Jeddah Reports Findings in Machine Learning (Smart deep learning model for enhanced IoT intrusion detection). Journal of Engineering. July 14, 2025; p 4450.