Intrusion detection using advanced salvation optimizer-based multi-head attention random multimodal deep learning model

Attacks are becoming increasingly common in network systems, and it is quite complicated to transfer trustworthy information over the network. Traditional intrusion detection models struggled with certain limitations, including limited performance, overfitting, false errors, and computational complexity. Recently, many deep learning (DL) techniques have been used to overcome the existing problems byt results computational issues. Hence, this research designs an advanced salvation optimizer-based multi-head attention random multimodal deep learning (ASO-MHA-MDL) model for intrusion detection, which detects multiple attacks over the network and overcomes these existing limitations. The ASO-MHA-MDL model enhanced the detection performance by adopting the metaheuristic optimization, the advanced salvation optimizer (ASO) technique used for feature selection and intrusion detection by parameter tuning, which makes it easier to train the model. The application of the MHA mechanism and advanced preprocessing techniques boosts the model’s structure to obtain accurate detection. These advancements increase the performance of the model for accurate intrusion detection over the network system with high detection accuracy while minimizing computational complexity. More specifically, the evaluation results show that the ASO-MHA-MDL model achieved a high accuracy of 96.85%, using the CSE-CIC-IDS2018 dataset and 97.12% using the UNSW-NB15 dataset, respectively, and outperformed other existing models..