An intrusion detection system in the Internet of Things with deep learning and an improved arithmetic optimization...

The Internet of Things (IoT) is vulnerable to cyber-attacks due to limited security mechanisms and resources constraints. Traditional intrusion detection systems (IDS) deal with imbalanced datasets, high-dimensional network traffic, and the inability to detect new attacks. This research proposes an advanced IDS framework that utilizes game-theory-based Generative Adversarial Networks (GAN) for dataset balancing, a hybrid Arithmetic Optimization Algorithm (AOA), and a Sine Cosine Algorithm (SCA) for feature selection, and a Parallel Convolutional Neural Network (PCNN) combined with Long Short-Term Memory (LSTM) layer for accurate attack detection. The suggested ASPCNNLSTM model achieves a precision of 99.86% on the NSL-KDD dataset and an attack detection accuracy of 98.67% on the UNSW-NB15 dataset, significantly outperforming traditional CNN, LSTM, and feature selection methods. This strategy improves IDS capability by selecting the best features relevant to traffic features and improving spatial and temporal feature learning, making it robust against complex and unknown cyber threats.