Network Intrusion Detection Using a Stacking of AI-driven Models with Sampling
Samar AboulEla, Rasha Kashef · 2024
Securing computer networks in Internet of Things (IoT) platforms has become increasingly challenging due to rising threats. Identifying unusual activities in IoT networks is crucial for maintaining their security. This research delves into detecting anomalies in network behavior to overcome security challenges. We use artificial intelligence (AI) models, specifically machine learning (ML) and deep learning (DL), for detecting intrusions in the network. We adopt four ML classifiers and implement two methods for assembling the classification outputs using stacking. Training our model on extensive historical network data enhances the ability to recognize abnormal network behaviors effectively. Initial tests on the NSL-KDD benchmark dataset have shown promising results, indicating the potential effectiveness of our approach. We also employed oversampling using Generative Adversarial Networks (GANs) to maintain balance in the data distribution, which led to noticeable improvement, reaching a 73.5% F-score and 61% accuracy compared to baseline models.