Predicting Different Types of Imbalanced Intrusion Activities Based on a Multi-Stage Deep Learning Approach

Intrusion Detection Systems for IoT networks have emerged to solve the vulnerabilities caused by the extensive utilization of IoT devices for different applications. Intrusion Detection Systems are not only limited to predicting the existence of intrusion activities apart from the normal ones but it is also extended to identify different types of intrusion activities that allow for a larger scale of recovery actions toward solving this security breach. This study proposes a deep learning approach to detect different types of intrusion activities using a multistage mechanism and an oversampling process which solves the problem of the imbalanced data produced by the IoT devices. This work confirms that the selected classification techniques are not able to detect all types of intrusion for imbalanced data if not combined with the oversampling process. It also compares the proposed approach with other classification and deep learning techniques which consider the oversampling process as part of their pre-processing phase. The results presented in this work show that the proposed approach outperforms the other techniques in terms of Accuracy and G-mean and has an advantage over the other techniques in predicting the different types of intrusion in terms of Precision and Recall.