Volume 17, Number 6

Advanced Intrusion Detection and Classification using Transfer Learning with Squeeze and Excitation Network and Adaptive Optimization in Big Data

  Authors

Viet H. Le ¹, Huy-Trung Nguyen ², Cuong V. Trinh ¹ and Tran Minh Hieu ², ¹ People’s Security Academy, Vietnam
² Research Institute of Posts and Telecommunications, Vietnam

  Abstract

domain have made cyberattacks, particularly Denial-of-Service (DoS) attacks, a very serious threat. Traditional intrusion detection systems face scaling issues with increasing complexity of big data, while extracting local and global features causes redundancy. This research bridges this gap by integrating transfer learning and Squeeze-and-Excitation Network (SENet) within the proposed eXplainable Artificial Intelligence-driven Intrusion Detection Model (XIIDM). The feature extraction process is through a correlated univariate-elimination-based autoencoder, to preserve local and global features of input data and eliminate all redundant information. SENet further enhances representational power of proposed model by recalibrating channel-wise feature responses, leading to improved DoS classification accuracy. An adaptive partial reinforcement optimizer dynamically adjusts model parameters during training, thus optimizing precision and reducing time complexity. Moreover, incorporation of explainable artificial intelligence units makes the outcome of XIIDM transparent and accountable. The proposed XIIDM is then rigorously evaluated on five benchmark datasets: CSE-CIC-IDS2018, CIC-DDoS2019, NSL-KDD, KDD Cup-99, and UNSW-NB15, achieving 99.988% accuracy, 99.934% precision, and 99.932% recall, with 0.00014% error rate. This research further justifies the robustness and generalization capacity of the proposed model by performing k-fold cross-validation and ablation experiments, confirming its high performance and reliability.

  Keywords

Intrusion detection systems, Denial-of-service attacks, Imbalanced dataset, Transfer learning, Cyber threats, Explainable artificial intelligence