Hybrid Temporal Convolutional and Anomaly Attention Network with Bi-Gru for Network Intrusion Detection

The complexity of evolving cyber attacks requires intrusion detection systems (IDS) capable of detecting long- and short-term attack patterns with real-time effectiveness. Existing solutions, however, cannot tackle class imbalance, high false pos itives, and computational overhead. To address these loopholes, we introduce TCANet, a hybrid deep learning model combining Temporal Convolutional Networks (TCN), Anomaly Attention Mechanisms, and Bidirectional GRUs (Bi-Gru). TCANet uses SMOTE oversampling and time-warping augmentation to tackle imbalanced data, recording 92% accuracy and 2.1% false positive rate (FPR) on the CIC-IDS-2018 dataset—20% faster inference than LSTMbaselines. This paper bridges the crucial gap between anomaly detection robustness and deployability in resource constrained settings.