With the rapid proliferation of the Internet, network security issues that threaten users have become increasingly severe, despite the widespread benefits of Internet access. Most existing intrusion detection systems (IDS) suffer from suboptimal performance due to data imbalance and feature redundancy, while also facing high computational complexity in areas such as feature selection and optimization. To address these challenges, this study proposes a novel network intrusion detection method based on an improved binary simulated annealing algorithm (IBSA) and TPE-FL-LightGBM. First, by integrating Focal Loss into the loss function of the LightGBM classifier, we introduce cost-sensitive learning, which effectively mitigates the impact of class imbalance on model performance and enhances the model’s ability to learn difficult-to-classify samples. Next, significant improvements are made to the simulated annealing algorithm, including adaptive adjustments of the initial temperature and Metropolis criterion, the incorporation of multi-neighborhood search strategies, and the integration of an S-shaped transfer function. These improvements enable the IBSA method to achieve efficient optimal feature selection with fewer iterations. Finally, the Tree-structured Parzen Estimator (TPE) algorithm is employed to optimize the structure of the FL-LightGBM classifier, further enhancing its performance. Through comprehensive visual analysis, ablation studies, and comparative experiments on the NSL-KDD and UNSW-NB15 datasets, the reliability of the proposed network intrusion detection method is validated.

Numerous studies attempt to mitigate classification bias caused by class imbalance. However, existing studies have yet to explore the collaborative optimization of imbalanced learning and model training. This constraint hinders further performance improvements. To bridge this gap, this study proposes a collaborative optimization Boosting model of multiclass imbalanced learning. This model is simple but effective by integrating the density factor and the confidence factor, this study designs a noise-resistant weight update mechanism and a dynamic sampling strategy. Rather than functioning as independent components, these modules are tightly integrated to orchestrate weight updates, sample region partitioning, and region-guided sampling. Thus, this study achieves the collaborative optimization of imbalanced learning and model training. Extensive experiments on 20 public imbalanced datasets demonstrate that the proposed model significantly outperforms eight state-of-the-art baselines. The code for the proposed model is available at: https://github.com/ChuantaoLi/DARG.