A Scalable Framework for Attack Detection: SWIM Transformer with Feature Optimization and Class Balancing

With the use of Internet-connected operating systems and the deployment of cutting-edge advanced knowledges like artificial intelligence (AI), Internet of Things (IoT), cloud computing, and deep learning within corporate organisations, industrial automation systems are undergoing a revolutionary transition. These new and creative ideas are helping to make Industry 4.0 possible. But these technological developments and the high-quality solutions they make possible will also bring up new security issues, the effects of which must be determined. Modern systems need sophisticated models that can handle big datasets, extract pertinent features, and deal with issues of class imbalance in order to identify attacks effectively. This paper presents a new framework that combines the SMOTE (Synthetic Minority Over-sampling Technique) model for balancing imbalanced datasets with the SWIM Transformer architecture and the Improved Pelican Optimiser for feature selection. By using attention mechanisms, the SWIM Transformer is able to precisely identify harmful activity by capturing complex patterns and correlations inside data. By improving feature selection, the Improved Pelican Optimiser lowers computing complexity while maintaining essential characteristics for precise classification. SMOTE also makes ensuring minority classes are fairly represented, which lessens the effect of unbalanced data on model performance. Results from experiments show that the suggested strategy outperforms current means in terms of detection accuracy, precision, recall, and F1-score. Additionally, the framework significantly decreased false positives and negatives, guaranteeing robustness and dependability in a range of assault scenarios. With possible uses in cybersecurity, intrusion detection systems, and other domains needing sophisticated anomaly detection, this research offers a scalable and effective attack detection method.