Explainable AI-Driven Intrusion Detection System for DoS Attack Classification Using Deep Learning and Optimization Techniques

Abstract

An Intrusion Detection System (IDS) is crucial for strengthening cyber resilience, particularly in detecting and mitigating Denial of Service (DoS) occurrences that compromise network operations. However, current IDS models face challenges such as class imbalance, redundant features, and a lack of interpretability. To address the above issues, this research proposes an Explainable Artificial Intelligence (XAI) driven IDS that integrates deep learning and optimization methods to improve the classification of DoS attacks. A reweighted auxiliary classifier generative adversarial network with gradient penalty is employed for data augmentation. An EfficientNet-B0 with hybrid attention modules enhances the feature extraction process, identifying critical patterns. A multiplicative Luong attention deep residual multiscale convolutional neural network is utilized to enhance the robustness and accuracy of network traffic sample categorization by extracting context features. The enhanced starfish optimization algorithm is employed to tune the hyperparameters of the proposed method, minimizing computational complexity while maximizing classification accuracy. To enhance the transparency and trust in model decisions, shapely additive explanations and local interpretable model-agnostic explanations are implemented, offering insight into feature importance, thereby improving both the transparency and reliability of the IDS. The proposed method is executed on five datasets: NSL-KDD, CSE-CIC-IDS2018, CCIDS2019, KDD-Cup99, and UNSW-NB15, attaining classification accuracies of 99.99%, 99.96%, 99.97%, 99.96%, and 99.95%, respectively, and it performs better than the current techniques.