Securing Networks with Convolutional Long Short-term Memory Based Traffic Prediction and Attention Mechanism for Intrusion Detection

In response to the escalating complexity of network environments and the heightened sophistication of cyber threats, the demand for robust and efficient Network Intrusion Detection Systems (NIDS) is now crucial. This paper presents an innovative NIDS approach utilizing Convolutional Long Short-Term Memory (ConvLSTM) networks and attention mechanisms for precise and timely intrusion detection. Our proposed model integrates ConvLSTM’s ability to capture spatiotemporal dependencies in network traffic data with attention mechanisms that enable the model to focus on pertinent information and filter out noise. We preprocess network traffic data into a sequential format, employ ConvLSTM layers to learn spatial and temporal features, and introduce an attention mechanism that assigns varying weights to different input data parts. This dynamic weighting emphasizes regions most likely to contain malicious activity. Extensive experiments on the CICIDS2017 Dataset validate the effectiveness of our approach, achieving an accuracy of 99.98%. This underscores its potential to revolutionize modern network intrusion detection and proactively safeguard digital assets.