Deep Learning Models with Transfer Learning and Ensemble for Enhancing Cybersecurity in IoT Use Cases

Internet of Things (IoT) applications have made inroads into different domains, providing unique solutions—Internet of Things technology offers seamless integration of physical and digital worlds. However, the broad nature of the technologies and protocols used in IoT applications has increased vulnerability from malicious attackers. Hence, protecting IoT applications from cyber-attacks is imperative. Researchers have implemented intrusion detection systems to overcome this issue to improve cybersecurity in IoT scenarios. With the new threats of cybercrime emerging, a continuous effort is required to enhance the security of IoT applications. To address this pressing need, we present our study that proposes a deep learning-based framework to bolster cybersecurity at the IoT use cases level by exploiting the power of transfer learning and ensembling it from deep learning models pre-trained at larger datasets. Deep learning models attain high performance with the help of hyperparameter tuning, and we achieve that through PSO in our proposed system. Our ensemble system shows how individual models can outperform individual models by using the best-performing models as constituents in the ensemble approach. We introduce an algorithm called — Optimized Ensemble Learning-Based Intrusion Detection (OEL-ID). This algorithm leverages the present framework and corresponding optimization strategies to boost intrusion detection performance for improved cyber security in IoT scenarios. Using the UNSW-NB15 benchmark dataset, our empirical study demonstrates that our proposed method, compared to some of the existing deep learning models, obtained a detection accuracy of 98.89%, which, in turn, provided the highest comparative accuracy. Therefore, the proposed system can be used with IoT use cases as it allows for a significant level of security to the system's underlying applications