Year: 2026
Type: preprint
Abstract: The Internet of Things (IoT)devices are spreading fast. This is happening at the time as new high-speed communication networks and 5G infrastructures are being built. Because of this there are a lot more Distributed Denial-of-Service (DDoS) attacks happening. These DDos attacks are also getting more complicated. This is a problem for the systems that are supposed to detect when someone is trying to break in. The DDoS attacks are a challenge, for these systems to deal with this problem. This research suggests a hybrid intrusion detection approach that combines Long Short-Term Memory (LSTM) networks with the Light Gradient Boosting Machine (LightGBM) classifier for efficient multi-class DDoS attack detection. The LSTM module is employed to capture temporal dependencies and sequential traffic patterns, while LightGBM performs fast and accurate classification based on the extracted feature representations. In addition, SHapley Additive exPlanations (SHAP) are incorporated to enhance the suggested framework's interpretability by determining the key elements influencing the detection process. we use in oure model tres dataset CICIDS2017, CIC-DIAD2024 and CICIoT2023. the results show that the framework achieves magnifique in multi-class classification performance, achieving accuracy of 97.80\%, and precision of 97.60\%, recall of 97.70\% and F1-score of97.60\% on the CIC-DIAD2024 dataset, achieve an precision 97.60\% ,accuracy of 96.80\%, recall 97.70\%, and F1-score of 97.60\% on CICIoT2023. Furthermore, the proposed model reaches an overall accuracy of 99.89\% on the CICIDS2017 benchmark dataset. The framework also maintains very low inference latency, making it useful for real time IoT intrusion detection. Experimental results confirme that the combination of gradient boosting for temporal feature learning greatly improves good detection accuracy, robustness, computational activity and the interpretability, and so on making the proposed framework a useful approach for improving the security of future IoT and 5G networks.
Source: Research Square
Institutions: University Yahia Fares of Medea, Gazi University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7204455834