The rapid proliferation of IoT devices has had a profound impact on various aspects of our daily lives, offering unparalleled convenience, automation, and enhanced productivity across a wide array of applications such as smart cities, healthcare, agriculture, and industries. However, this exponential growth of IoT gadgets has also led to a corresponding increase in security threats and cyber-attacks. Therefore, a robust and effective smart intrusion detection and prevention system (IDPS) hinges on a learning algorithm is required for securing these gadgets from cyber-attacks. In this paper, an IDPS for the smart IoT system is proposed. The initial contribution of this study involves introducing the modified grey wolf optimization algorithm with Monte Carlo and Levy Function (M-GWO-MC-LF). This algorithm is proposed for the hyper-parameter optimization of the artificial neural network (ANN). By optimizing the hyper-parameters of the ANN classifier, notable improvements are achieved in terms of accuracy and training time. Subsequently, we present a prevention approach aimed at mitigating invasive traffic by identifying and blocking source IP addresses. This approach is implemented by placing the entire IDPS behind the end-user's firewall, ensuring that potential intrusions are effectively prevented. To assess the efficacy of the proposed system, experiments are carried out utilizing both the BoT-IoT dataset and real-time network traffic data.