Attentive transformer deep learning algorithm for intrusion detection on IoT systems using automatic Xplainable...

Area of Science:

• Cybersecurity
• Machine Learning
• Internet of Things (IoT) Security

Background:

• The proliferation of Internet of Things (IoT) and Industrial Internet of Things (IIoT) systems increases security risks, particularly from denial of service (DoS) and distributed denial of service (DDoS) attacks.
• Existing intrusion detection systems (IDS) often struggle with the tabular data format common in machine learning tasks and face challenges in model explainability and feature selection.

Purpose of the Study:

• To propose an intelligent intrusion detection system (IDS) for IoT security that addresses the limitations of traditional deep learning models on tabular data.
• To develop a model that utilizes attentive mechanisms for automatic salient feature selection and provides explainable results.

Main Methods:

• Implementation of the TabNet-IDS model using the TabNet algorithm within the PyTorch deep-learning framework.
• Utilizing attentive mechanisms within TabNet for automatic feature selection and enhanced model interpretability.
• Evaluation of the model's performance on benchmark datasets: CIC-IDS2017, CSE-CICIDS2018, and CIC-DDoS2019.

Main Results:

• The TabNet-IDS model achieved high accuracy rates on the tested datasets: 97% on CIC-IDS2017, 95% on CSE-CICIDS2018, and 98% on CIC-DDoS2019.
• Demonstrated the effectiveness of the TabNet architecture for intrusion detection in IoT environments using tabular data.
• The model's attentive mechanisms provided explainable feature selection, contributing to better model understanding.

Conclusions:

• TabNet is a viable and effective deep learning architecture for intrusion detection on tabular datasets in IoT security.
• The proposed TabNet-IDS offers a robust solution for enhancing IoT security by accurately detecting network threats with explainable insights.
• Future work can explore further optimizations and applications of attention-based models for advanced cybersecurity challenges.