A Deep Learning-Based Intrusion Detection System for MQTT Enabled IoT | JoVE Visualize

A Deep Learning-Based Intrusion Detection System for MQTT Enabled IoT

Muhammad Almas Khan¹, Muazzam A Khan¹, Sana Ullah Jan²

¹Department of Computer Sciences, Quaid-i-Azam University, Islamabad 44000, Pakistan.

Sensors (Basel, Switzerland)|November 13, 2021

View abstract on PubMed

Summary

A Deep Neural Network (DNN) effectively detects intrusions in Message Queuing Telemetry Transport (MQTT) communications. This DNN model outperforms traditional machine learning algorithms, enhancing IoT security against various cyber threats.

Area of Science:

Cybersecurity
Network Protocols
Machine Learning

Background:

Internet of Things (IoT) devices utilize diverse messaging protocols, with Message Queuing Telemetry Transport (MQTT) being prevalent for sensor data.
The publish-subscribe model of MQTT, while efficient, presents vulnerabilities exploitable by attackers.
Intrusion detection is critical for securing IoT environments against evolving cyber threats.

Purpose of the Study:

To propose and evaluate a Deep Neural Network (DNN) for intrusion detection specifically within MQTT-based communication protocols.
To compare the performance of the proposed DNN model against traditional machine learning (ML) algorithms.

Main Methods:

A Deep Neural Network (DNN) model was developed for intrusion detection in MQTT.
The DNN's performance was evaluated against Naive Bayes (NB), Random Forest (RF), k-Nearest Neighbour (kNN), Decision Tree (DT), Long Short-Term Memory (LSTM), and Gated Recurrent Units (GRUs).
Experiments were conducted using two datasets: MQTT-IoT-IDS2020 and a dataset featuring Man in the Middle (MitM), network intrusion, and Denial of Services (DoS) attacks.

Main Results:

For the MQTT-IoT-IDS2020 dataset (binary classification), the DNN achieved high accuracies: 99.92% (Uni-Flow), 99.75% (Bi-Flow), and 94.94% (Packet-Flow).
In multi-label classification on the same dataset, accuracies were 97.08% (Uni-Flow), 98.12% (Bi-Flow), and 90.79% (Packet-Flow).
Against LSTM and GRUs on the second dataset, the DNN model achieved a leading accuracy of 97.13%.

Conclusions:

The proposed DNN model demonstrates superior performance in detecting intrusions within MQTT-based IoT environments.
DNNs offer a robust solution for enhancing the security of MQTT communications compared to traditional ML methods.
The findings highlight the potential of deep learning for safeguarding IoT networks against sophisticated cyberattacks.