Enhanced Network Intrusion Detection Using a Hybrid CNN-LSTM Approach on the UNSW-NB15 Dataset

In the era of rapid technological advancements, data and information have become significant assets that require continuous protection from unauthorized access and malicious attacks. Network Intrusion Detection Systems (NIDS) can effectively manage the increasing data demands assuring robust security. However, traditional or manual methods are often inadequate to cope with the sophisticated and evolving nature of cyber threats due to their inability to quickly process large volumes of data. Traditional machine learning models are widely used for this purpose. Recently deep learning approaches have become popular in classification tasks showing promising results. This paper introduces a hybrid deep-learning approach for detecting various network intrusions using binary and multi-class (ten classes) classification on the well-known UNSW-NB15 dataset. The proposed hybrid CNN-LSTM model demonstrates significant improvements in intrusion detection, achieving an accuracy of 97.19% for binary classification and 87.70% for multi-class classification, particularly in handling complex attack patterns and reducing false positives. Evaluated using metrics such as precision, recall and F1-score, the model consistently performs robustly across nine distinct attack categories. These results highlight the model's effectiveness in establishing enhanced network security.