The increasing complexity of network environments poses significant security challenges, particularly in defending against Denial of Service (DoS) attacks. Traditional security solutions are often inadequate, necessitating the development of intelligent procedures. This paper explores innovative approaches using deep learning-based Intrusion Detection Systems (IDS), specifically Bidirectional Long Short-Term Memory (BiLSTM) models, for enhanced attack detection. Our findings demonstrate that BiLSTM models excel in binary classification and outperform in identifying sophisticated attacks within multiclass settings. Two key areas for future research are highlighted: the impact of advanced data processing techniques on IDS effectiveness and the evaluation of Distributed Denial of Service (DDoS) attacks. The study introduces a refined BiLSTM (RLSTM)-based IDS approach and evaluates its performance using datasets CICIDS-2019, CICIDS-2017, and NSL-KDD. The method incorporates preprocessing techniques such as encoding, dimensionality reduction, and normalization. Experimental results indicate high accuracy, with 99.2% for NSL-KDD, 99.4% for CICIDS-2017, and 99.7% for CICIDS-2019 datasets. These results illustrate that the proposed technique provides superior detection capabilities in complex network environments compared to existing methods. This study emphasizes the necessity for advanced security measures and highlights the effectiveness of model in enhancing network protection. • Proposes deep learning–based IDS using a refined BiLSTM (RLSTM) to counter DoS attacks in complex networks. • Achieves high detection performance in both binary and multiclass attack scenarios. • Integrates effective preprocessing techniques, including encoding, dimensionality reduction, and normalization. • Validated on NSL-KDD, CICIDS-2017, and CICIDS-2019 datasets with accuracies up to 99.6%. • Demonstrates superior robustness and detection capability compared to existing IDS approaches.
M et al. (Sun,) studied this question.