A multiplicative attention-based deep learning framework for intrusion detection in IIoT networks

The Industrial Internet of Things (IIoT) has become increasingly popular. However, this growth has also introduced more vulnerabilities to various services and systems, making them prime targets for attacks aimed at stealing sensitive data. Therefore, IIoT network security is an important factor for which the intrusion detection system (IDS) is often used to monitor network traffic and identify malicious activities, where machine learning (ML) and deep learning (DL) algorithms play an important role. Generally, the existing systems detect a limited number of attack classes and face challenges in identifying attack sub-categories. As the number of attack classes increases and the training data becomes highly imbalanced, the performance of these systems decreases. To overcome these challenges, this paper proposes a multiplicative attentionbased deep learning framework for predicting malicious attacks in IIoT networks. The proposed model consists of multiplicative attention to compute the importance of each input feature. The convolutional layers are used to filter these parameters and a feed-forward neural network (FFNN) is used to make predictions. To assess the performance of the proposed framework, we utilized the CICIoT2023 dataset, which contains highly imbalanced data for attack categories and sub-categories. The proposed approach achieved 94.61% accuracy in category classification and 90.7% accuracy in sub-category attack classification.