As the complexity of cyber-attacks and network environment grow, we need more sophisticated ways for network intrusion detection. Long Short-Term Memory (LSTM) networks were applied to investigate the improvement of network intrusion detection system (IDS) performance. We evaluate the performance of LSTM against classic machine learning algorithms (i.e., Decision Trees (DT), Random Forest (RF), K-Nearest Neighbors (KNN), and Support Vector Machines (SVM)) and assess these algorithms with methodology such as accuracy, precision, recall, and F1 score. We show that LSTM does better than other classifiers and reaches an accuracy of 98.2% over other classifiers. By applying deep learning techniques to enhance multi-class network intrusions detection, this research advances on the improvement of more effective IDS. At the same time, the current work also highlights the need for feature reduction strategies that can enhance model performance and flexibility further. Future work will involve using larger datasets such as UNSW-NB15 and easing detection of evolving attack strategies. This study makes two contributions: first, this is the first study that comprehensively evaluates deep learning in network security, laying a solid foundation for future improvement of the IDS.
