An Intrusion Detection System (IDS) is crucial for safeguarding computational resources and data from external attacks on computer networks. However, modern IDSs face challenges in enhancing their resilience and reliability, particularly when dealing with unexpected and unpredictable threats. Deep neural networks are widely recognized as a powerful machine learning approach for handling complex systems with abstract properties. In this research, we propose a deep learning methodology utilizing a Long Short-Term Memory (LSTM) to develop an accurate and scalable IDS. LSTM has demonstrated exceptional performance in tasks such as voice recognition in speech detection and can also be effectively applied to supervise learning on time series data. We designed a machine learning model based on LSTM by encoding Transmission Control Protocol/Internet Protocol (TCP/IP) packets within a fixed time window, simulating intrusion traffic patterns for the IDS. The model categorizes and labels normal and anomalous network traffic, enabling supervised learning with LSTM. To validate the efficiency of our model, we conducted experiments using the UNSW NB15 IDS dataset. The models were trained over 200 epochs with a learning rate of 0.001 on both balanced and imbalanced data. Results showed that the LSTM and its architectural variations outperformed classical machine learning classifiers, largely because of the LSTM’s ability to capture high-level abstract features from low-level network traffic characteristics.