SHO-Xception: An Optimized Deep Learning Framework for Intelligent Intrusion Detection in Network Environments

Advanced, efficient, and scalable Intrusion Detection Systems (IDS) are required due to rise of cyber threats in current digital infrastructures. By combining Z-score normalisation for preprocessing, a Sea-Horse Optimiser (SHO) for feature selection, and Xception architecture for classification, this study launches a new framework for IDS that is based on deep learning. suggested model improves learning performance with little parameter overhead by using Xception's residual connections and depthwise separable convolutions, which are computationally advantageous. benchmark dataset for evaluation is UNSW-NB15 dataset, which contains a rich mixture of normal and attack traffic across ten classes. To improve model generalisability and training speed, SHO is used to choose most relevant and high-impact features. To thoroughly test model in both binary and multiclass classification scenarios to see how it performs. Achieving high per-class F1-scores in multiclass scenarios and an average accuracy of 95.6% in binary classification, experimental results show that suggested SHO-Xception model performs superiorly. discriminative capacity of model is further validated via ROC curves and AUC values. suggested model outperforms state-of-the-art methods in recall, robustness, and precision when compared to CNN+PSO, LSTM+GWO, ResNet+ACO, and SVM+RFE. For next-gen intrusion detection system (IDS) applications in complicated network settings, suggested framework demonstrates a realistic and scalable approach