Adaptive GraphSAGE-DQN Routing for Energy- Efficient Wireless Sensor Networks

Abstract

Wireless Sensor Networks (WSNs) are important in the contemporary communication systems as they facilitate effective data gathering and transmission where there is scarcity of resources. Nonetheless, issues like energy inefficiency, dynamic network topology and susceptibility to malicious attacks have a serious impact on worsening network performance and lifetime. To solve these problems, this paper suggests a hybridized Graph Neural Network-Deep Q-Network (GNN-DQN) model to make intelligent and adaptive decisions in WSNs. The model under consideration combines a Graph Neural Network (GNN) to represent spatial and topological correlations between sensor nodes with a Deep Q-Network (DQN) to optimize the policies of routing and classification, depending on the energy usage, transmission patterns, and network condition. The GNN component captures the high-level node embedding on the basis of inter-node relationships and the DQN component uses the high-level node embedding to optimize action-value in order to make an efficient decision under dynamic environments. The model is tested on the publicly available dataset WSN-DS and it is based on extensive preprocessing, feature engineering, and normalization. It is experimentally shown that the proposed method has high classification accuracy (up to 98-99%), better F1-score, and high robustness in numerous classes. Moreover, the model can outperform all the traditional machine learning models like the Logistic Regression and XGBoost, and even deep learning baselines. The assessment of the performance based on the parameters of accuracy, F1-score, sensitivity, ROC-AUC, and Precision-Recall curves proves that the proposed framework is effective to improve detection potential and decision efficiency. The findings suggest that graph-based learning combined with the reinforcement learning is an effective and scalable method to intelligent WSN management.