ATTENTION-ENHANCED HYBRID DEEP NEURAL NETWORK FOR INTRUSION DETECTION ACROSS HETEROGENEOUS NETWORK DATASETS

Abstract

Intrusion detection has become a critical component of modern cybersecurity infrastructures as network environments continue to expand in scale and complexity. Conventional intrusion detection systems often rely on machine learning models trained on a single dataset, which limits their ability to generalize across heterogeneous network environments. This lack of cross-dataset adaptability leads to reduced detection accuracy and increased false alarms when deployed in real-world systems where traffic patterns vary significantly. To address this limitation, this study proposes an Attention-Enhanced Hybrid Deep Neural Network designed to improve intrusion detection performance across heterogeneous network datasets. The proposed framework integrates convolutional neural networks for hierarchical feature extraction, a bidirectional long short-term memory module for temporal dependency modeling, and an attention mechanism that selectively emphasizes informative traffic features. Experiments are conducted using three widely adopted benchmark datasets: NSL-KDD, UNSW-NB15, and CICIDS2017. The model is implemented using Python with TensorFlow and evaluated using standard performance metrics including accuracy, precision, recall, and F1-score. Experimental results demonstrate that the proposed model achieves an average detection accuracy of 98.3%, outperforming several existing deep learning models while maintaining low false positive rates. The findings highlight the effectiveness of attention-guided hybrid learning for robust intrusion detection across heterogeneous network environments.