A multi-objective bayesian optimized neural network approach for intrusion detection

Cyberattacks in Internet of Things (IoT) and vehicular contexts are becoming more sophisticated, which in turn emphasises the need for an ideal Intrusion Detection System (IDS) that provides high detection accuracy and computational efficiency. This research proposes MambaEcoNet-SC1D, a novel IDS model that encompasses multi-stage feature learning with intelligent optimization. In this multi-stage pipeline, Sparse CNN-1D first extracts compact local traffic features, the Eco Transformer captures contextual dependencies through energy-efficient distance-based attention, and the SSM-based Mamba block models long-term temporal patterns, collectively improving detection accuracy, robustness, and computational efficiency. The architecture combines Sparse CNN-1D for compact local feature extraction, an Eco Transformer block employing distance-based Eco Attention for efficient contextual encoding and an SSM Mamba block for scalable long-sequence modelling. Furthermore, the hyperparameters of the model are simultaneously adjusted using Multi-Objective Bayesian Optimisation (MOBO), which guarantees balanced performance in terms of detection accuracy, false alarm rate and inference efficiency. Extensive experiments on two benchmark datasets validate the effectiveness of the proposed model. On CICDDoS-2019, MambaEcoNet-SC1D achieves 98.87% accuracy, 98.85% precision and 98.86% F1-score, while on CSE-CIC-IDS2018, it delivers near-perfect results with 99.99% across all core metrics. A comparative evaluation against recent IDS frameworks further demonstrates the superiority of the proposed model, achieving consistent improvements in accuracy and specificity.