AttMT-ConNet: A Lightweight and Efficient Intrusion Detection Model for Edge Computing

Abstract

A lightweight, energy-efficient model-based intrusion detection is a security mechanism that employs deep learning (DL) methods in order to detect attacks in the network or IoT environment using minimum memory and minimum energy consumption. However, the conventional intrusion detection systems (IDS) are complex, huge in size, and consume high energy levels, which makes them inefficient for use in edge devices having small computing abilities. To overcome these limitations, a lightweight and energy-efficient intrusion detection model is proposed using a DL approach for resource-constrained edge computing environments. Initially, the input data is obtained from the CIC-IDS-2017 network intrusion dataset for validating the system performance. Also, Tail Robust Quantile Normalization (TRQN) is used to deal with outliers and skewness in order to improve the quality of data and make the system more robust. Besides that, the Mutual Information-Interaction Gain based Feature Selection (MI-IGFS) model is employed to extract the most important features, and thus decrease the model size and computational cost. For detecting intrusions, an Attention-based Multi-scale Temporal Convolutional Network (AttMT-ConNet) is developed to effectively capture temporal patterns and multi-scale dependencies in network traffic data. Incorporating attention mechanisms enhances feature representation and detection accuracy while keeping efficiency constant. The architecture implementation is carried out using the Python programming language, and performance measures are evaluated. It proves to be efficient for scalable, secure, and real-time intrusion detection in the edge computing environment.