Wireless sensor networks (WSNs) are designed with small and energy-efficient sensor nodes (SNs), which are made up of microcontroller and sensors, are a critical module, which is used to collect data and to perform computations. Numerous industries include these networks, for a variety of domains like health care, monitoring, eco-friendly, business, and martial operations. Though Wireless Sensor Networks (WSNs) are widely used due to their ability to exchange large volumes of data, their extensive usage presents unique challenges. Sensor nodes (SNs), which are the basic building blocks of WSNs, play a crucial role in addressing these challenges and ensuring efficient data transmission. To preclude information breaches and failure of network, security measures are crucial. Using Artificial Intelligence (AI) methods, such as Decision Tree (DT), Support Vector Machine (SVMs), Naïve Bayes (NB), and Random Forest (RF), for detection malicious nodes based on difference between predicted and real sensor readings is a good method to increase the security of WSN. In the real-time applications, SNs aspect for anomalies, and the base station performs classification using ML methods. Conversely, existing methods have limitations which include high false positive rate and inadequate sensitivity to detect offensive patterns. Therefore, the authors deploy a Graph Neural Network (GNN)-based ML approach to detect malicious nodes in order to reduce these faults and enhance the Confidentiality, Integrity, and Availability (CIA) of networks. The proposed model revealed the outcomes, which have an accuracy rate of 98% and 98% for precision, recall, and F1 support. The model executes better than prior approaches, allowing to a relative investigation, which marks an important improvement in the security of WSNs in a diverse of applications.