<p>Recently, with the rapid advancement of the internet and communication technologies, the security of the network system has become more significant. Intrusion Detection Systems (IDS) are essential for detecting and mitigating unauthorized access, and protecting the information from potential threats. Accurate intrusion detection is mandatory for enhancing network security and preventing data from unauthorized access. Prior, various Intrusion Detection System based methodologies were developed to detect the intrusion that affects network security. Yet, they have struggled with poor detection performance and high computational complexity. This study proposes a novel Capsule Long Short-Term Memory-based Transformer Network for accurately detecting the intrusion. In this research, the Oppositional Levy-based Pelican Optimization is employed for optimal feature selection that reduces the system complexity, and irrelevant features and noise. Additionally, the temporal-self-attention mechanism is employed to capture the important patterns and dependencies within the time series data. Further, the ResNet-Convolution LSTM is employed to capture the long as well as short-term dependencies and reduce the vanishing gradient troubles. The performance of the Capsule Long Short-Term Memory-based Transformer Network is validated using UNSW-NB15, ToN_IoT, and, BoT-IoT datasets and compared to previous attack detection mechanisms in terms of some common assessing metrics. The experimental validation demonstrates that the Capsule Long Short-Term Memory-based Transformer Network effectively detected the intrusion and achieved a high accuracy of 98.89%, a high detection rate of 98.39%, and less false alarm rate of 2.85% compared to existing methodologies that underscoring its effectiveness in network security enhancement.</p>