A Trust-Weighted Stacking Ensemble Framework for Wormhole Attack Detection in Wireless Sensor Networks

Abstract

Wireless Sensor Networks (WSNs) are extensively employed in smart monitoring, healthcare, industrial automation and agriculture. However, due to the open communication environment and limited resources, they are quite vulnerable to routing attacks such as wormhole attacks. Current detection approaches are primarily based on hardware support, topological assumptions, or single-model learning techniques, which impair their scalability and adaptability. In this paper, we present a novel architecture called Adaptive Trust-Weighted Ensemble with Dynamic Feature Fusion (ATWEDF) for wormhole attack detection in WSNs. The proposed approach uses classifiers such as Random Forest, XGBoost and Support Vector Machine to provide behavioural, communication and verifier trust signals. A correlation-based dynamic feature fusion mechanism is utilized to encourage discriminative feature representation and Logistic Regression meta-learner performs adaptive trust-weighted stacking for final classification. The studies have been performed on Wormhole attack-Contr2 v2 dataset which consists of 637,862 records with 20 input features. The experimental findings reveal that ATWEDF obtains 99.75%, 99.83%, 99.84%, 99.83% and 0.55% for Accuracy, Precision, Recall, F1-score and False Positive Rate respectively which is better than the existing baseline models. The results confirm that trust-weighted stacking with correlation-driven feature fusion is an effective and reliable solution for wormhole attack detection in WSNs.