Enhancing Intrusion Detection Through Data Perturbation Augmentation Strategy

Intrusion data augmentation is an approach used to increase the size of the training data sample to improve the classification capabilities of machine-learning algorithms applied to intrusion detection. In this study, we introduced data perturbation by adding Gaussian noise to the minority class representing the intrusion scenarios. Employing the Divide-Sort, Augment, and Combined (SAC) technique, we performed oversampling on the minority class of two datasets used for training the model. Subsequently, we validated the model to achieve high overall accuracy indicating reliable intrusion detection. The performance of the model on the perturbed dataset was compared with that of the SMOTE and ROSE data augmentation methods. The results revealed that the perturbation of oversampled data exhibited superior and near perfect classification compared with the SMOTE and ROSE data augmentation techniques. The effectiveness of the proposed intrusion detection approach has been demonstrated on the BoT-IoT and smart grid imbalanced datasets, previously used for benchmarking.