ENHANCING IOT/IIOT INTRUSION DETECTION: A COMPARATIVE STUDY OF HYBRID CNN-LSTM AND ADVANCED DNN ML MODEL ON EDGE-IIOTSET

Abstract

In this paper, we present an improved hybrid CNN-LSTM model to improve intrusion detection in IoT/IIoT environments. It is evaluated on the publicly available Edge-IIoTset dataset (hosted on Kaggle). Due to storage constraints, we use a dense but representative subset of the dataset, preserving its diversity in terms of attack scenarios and device interactions. The original dataset covers a lot of things about IoT/IIoT devices (e.g., temperature/humidity sensors, ultrasonic sensors, flame detectors, and heart rate monitors) and includes 14 different attack types, including DDoS/Distributed Denial of Service (DDoS), Man in the Middle attack, MID, MID, attack, and spy attacks.

Our hybrid CNN-LSTM model uses the features of convolutional neural networks (CNN) for temporal pattern recognition and long term short term memory (LSTM) networks to simulate the network as a temporal encoder, as deep neural networks (DNN) are used in previous work. Through extensive experiments, we show that the model exhibits good accuracy, good classification accuracy, and very low overhead in the classification of the model. We also present a comparative study of the design tasks, demonstrating the model’s ability to establish the edge in low-end IoT environments.

The results show that the CNN-LSTM hybrid architecture is more robust against multiple attack vectors, thus providing predictive and robust solutions to real-time threats. This study helps to the development and study of intelligent IoT/IIoT-based computer security systems with applications in healthcare, medicine, and critical infrastructure. Data sharing and interoperability are key factors to improve research and studies in the field of IoT cybersecurity.