Dynamic channel alignment and temporal context attention fusion network for intrusion detection

Modern information systems are seriously threatened by the growing sophistication and number of network-based cyberattacks, necessitating sophisticated intrusion detection systems with real-time, high-accuracy classification capabilities. Conventional techniques frequently struggle to capture temporal dependencies in sequential network flows and inter-feature correlations among network traffic attributes. Motivated by this challenge, this research proposes a novel dual-branch Channel Dynamic Alignment and Temporal Context Attention Fusion Network (CDTA-FusionNet) model for high-performance intrusion detection on the cybersecurity intrusion detection dataset. The CDTA-FusionNet is a dual-branch integrating Channel Dynamic Residual Network (CDResNet) One Dimensional (1D) and Temporal-Attention Long Short-Term Memory (TA-LSTM) for network attack classification. The CDResNet leverages a residual block integrating Channel Spatial Excitation (CSE) and Dynamic Feature Alignment (DFA) modules to extract the spatial context feature vector from normalized traffic sequences. The TA-LSTM encodes the temporal dependencies through combining an attention layer with a Temporal Gradient Gating (TGG) module to extract a context temporal feature vector to emphasize sudden deviations indicative of attacks. After concatenating and processing the outputs from the two branches using dense fusion layers with ReLU activations, dropout, and batch normalization, classifying the binary attack with a sigmoid output layer. The preprocessing pipeline produces input shapes of samples, sequence length, and features appropriate for CDTA-FusionNet processing by managing missing values, label encoding categorical features, normalizing numerical features, and generating sliding-window sequences. According to experimental evaluation, CDTA-FusionNet outperforms traditional CNN, LSTM, and hybrid architectures with 99.98% classification accuracy. These contributions establish CDTA-FusionNet as a very efficient framework for cybersecurity intrusion detection, offering network traffic analysis feature representations that are both highly accurate and comprehensible.