An Effective and Efficient Intrusion Detection System of Network Attacks Using Stacked CNN and Voting Technique

Abstract

IDS are crucial to network security because they can identify malicious activity and halt it in its tracks. Network intrusion data is often masked by a sea of benign data, making it difficult to train a model or perform a detection with a high FPR. This is because networks are inherently dynamic and change over time. In this research, we offer a ML & DL model-based method to ID, and we demonstrate how to deal with the issue of data imbalance by using a hybrid sampling technique. Conventional firewalls and data encryption technologies are unable to provide the level of security required by current networks. As a result, IDSs have been endorsed for use against network threats. Recent mainstream ID approaches have benefited from ML, but they have low detection rates & need a lot of feature engineering to be truly useful. Using layered CNN and Voting classifier (XGBoost and LGBM), this study introduces ML-DL-NIDS to address the issue of subpar detection precision. Using a publicly available NSL-KDD & UNSW-15 benchmark datasets for network intrusion detection, we find that this model outperforms competing methods according to accuracy and F1-score obtained from experimental evaluations.