<![CDATA[The rapid expansion of cloud computing, Internet of Things (IoT), and distributed network environments has significantly increased vulnerability to sophisticated cyber threats, exposing the limitations of traditional signature-based intrusion detection systems. Although deep learning techniques, particularly Convolutional Neural Networks (CNNs), have shown promising performance in intrusion detection, challenges related to validation transparency, statistical reliability, and interpretability remain inadequately addressed. This study proposes an intelligent CNN-based intrusion detection framework designed to improve detection accuracy, robustness, and model explainability. The framework is evaluated using the UNSW-NB15 benchmark dataset, which reflects realistic modern cyber-attack scenarios. A comprehensive preprocessing pipeline involving data cleaning, categorical encoding, feature normalization, and data reshaping is applied to enhance learning efficiency. To ensure unbiased evaluation, stratified k-fold cross-validation and an independent held-out test set are employed. Experimental results demonstrate that the proposed CNN achieves a test accuracy of 91.8%, with balanced precision, recall, and F1-score across benign and malicious traffic classes. Multi-class detection analysis further confirms the model’s capability to distinguish among diverse attack categories. Statistical validation using mean performance metrics, standard deviation, and confidence intervals demonstrates stable generalization performance. Additionally, Gradient-weighted Class Activation Mapping (Grad-CAM) is used to enhance interpretability by identifying network-level features that influence classification decisions. An ablation study further validates the effectiveness of key architectural components. The results indicate that the proposed framework provides a reliable, scalable, and interpretable solution for intelligent intrusion detection in modern high-dimensional network environments.]]>
