The increasing number of Internet users has made security a crucial aspect of life in the modern Internet world. Numerous researchers have created a variety of intrusion detection systems in the past to use data analyzing techniques to identify and detect intrusions. Although traditional intrusion detection technology has been extensively studied for many years, its real-time detection performance may be insufficient. Designing a detection technique to quickly identify the threats is therefore essential. However, when utilizing analyzing data, the current methods are unable to attain a suitable level of detection accuracy. In order to overcome these drawbacks, Cosine Convolutional Graph Neural Network with Crested Porcupine Optimizer (CCGNN-CPO) method for intrusion detection in wireless networks is proposed in this research. In this input data is taken from two datasets, such as WSN-DS and CICIDS-2017 datasets. To reduce noise in these input data’s, Min–Max Normalization (Min-MaxN)-based preprocessing is proposed. Following that, the preprocessed data undergo feature selection using Poplar Optimization Algorithm (POA). After that classification is done using Cosine Convolutional Graph Neural Network (CCGNN) and optimized using Crested Porcupine Optimizer (CPO) for detect intrusion demonstrating superior efficiency and accuracy. The testing results demonstrate 99.1% accuracy in recognizing various threats, making them more accurate than existing algorithms. The outcomes of the proposed technique showed that it could improve reliability, real-time threat detection, and the evaluations ability of the intrusion detection in wireless networks.