Network Intrusion Detection Using Multi-Modal Ensemble Modeling

In recent times, networks and related data have greatly increased because of quick developments in the internet and communication domains. As a result, we witness a surge of new verities of security threats, which has made it difficult for network security professionals to identify breaches and mitigate them promptly. Moreover, it is impossible to overlook the existence of hackers who intend to carry out different types of assaults on the network. One such technology is an intrusion detection system (IDS), which guards against potential network invasions by examining network traffic to guarantee its availability, confidentiality, and integrity. There have been several works in this regard where we have advanced from traditional tools to Machine learning-based techniques. In this paper, we demonstrate the strength of ensemble modelling using a linear weighted combination of XGBoost and CNN. We specifically aim at utilizing network logs in both their tabular form (with XGBoost) as well as in the form of images (with CNN) for utilizing the data to its full potential. We compare our performance with existing benchmarks and observe that our proposed method outperforms them. Our proposed method achieves a high Recall and Precision, with an overall F1-score of 0.995.