Intrusion Detection System in IoT 5G Networks Based on LSSVM and Harmony Search Optimization

Abstract

5G-powered Internet of Things devices, there has been a major challenge in making its network infrastructures safe against the rising wave of cyber threats. In the context of 5G-IoT networks, the traditional intrusion detection systems (IDS) tend to have problems with real-time detection, class imbalance, and adaptive patterns of the attacks. In this study, a new hybrid system of Least Squares Support Vector Machine (LSSVM) and the Harmony Search Optimization Algorithm is proposed as a new intrusion detection framework capable of improving sensitivity and stable intrusion detection. Also, Principal Component Analysis (PCA) is used to decrease the feature dimensionality and get rid of the redundancy. The suggested model is tested with the use of reasonable botnet and adversarial traffic circumstances in the IoT-23 dataset. As far as the detection of such underrepresented attacks as U2R/R2L is concerned, SMOTE is used to balance the classes. Results demonstrate that the LSSVM + HSOA model achieves superior detection performance with an accuracy of 99.27%, significantly outperforming standard SVM and Random Forest baselines. The framework also shows improved recall for minority attack classes, affirming its suitability for complex and imbalanced IoT traffic. Future work will address real-time deployment challenges, such as latency and adversarial evasion, through lightweight model adaptations and distributed learning. This study contributes a practical and scalable approach to securing modern 5G-IoT networks.