Intrusion Detection Systems (IDSs) play a vital role in identifying and mitigating malicious network activities and system misuse. The integration of Artificial Intelligence (AI), particularly Deep Learning (DL), has significantly enhanced the adaptability and efficiency of IDS. This paper proposes an intelligent network-based IDS leveraging a DL model trained on the CSE-CIC-IDS2018 dataset. Key data pre-processing steps included duplicate removal, handling missing values, conversion of categorical data to a numerical form, and feature scaling. Initially, the model aimed to classify all individual attack types alongside benign traffic; however, the frequent misclassification of certain attack types prompted the aggregation of similar attacks into broader categories. This adjustment led to notable improvements in the performance metrics, including accuracy, precision, recall, and F1-score. To mitigate overfitting, weight decay in the context of neural networks, known as L2 weight regularization, was applied. The proposed improved DL model achieved an accuracy of 99.91%, precision of 98.61%, recall of 93.18%, and an F1-score of 94.78%, highlighting both the robustness of DL in intrusion detection and the critical role of comprehensive data preprocessing.
Bouidaine et al. (Sat,) studied this question.