An Intelligent Role-Based Intrusion Detection System using Hybrid Voting Classifiers for Securing Wireless Sensor Networks

Abstract

Wireless Sensor Networks (WSNs) play a crucial role in modern communication systems by enabling data collection and monitoring in applications such as environmental monitoring, healthcare systems, military surveillance, and smart cities. These networks consist of distributed sensor nodes that communicate with each other and transmit data to a base station. However, due to their open communication environment and limited security mechanisms, WSNs are highly vulnerable to routing attacks such as Flooding, Time division Multiple Access (TDMA) manipulation, Gray hole, and Blackhole attacks. These attacks can disrupt data transmission, increase energy consumption, and significantly reduce the reliability and lifetime of the network. Traditional intrusion detection approaches in WSN rely on rule-based or signature-based techniques that analyse predefined patterns of malicious behaviour. While these methods are useful for detecting known attack types, they often fail to identify new or complex attack patterns. To address these challenges, this study proposes a machine learning based routing attack detection system that analyses network communication parameters to identify malicious activities. Several machine learning classifiers including Decision Tree (DTC), Ridge Classifier (RC), and Linear Discriminant Analysis (LDA) are implemented as baseline models to evaluate routing attack detection performance. The proposed system introduces a hybrid classification model that combines Echo State Network (ESN) and Decision Tree Cost Complexity pruning (DTCCP) classifier also known as Deep Reservoir Routing Defense Model (DRRD) to improve attack detection accuracy. The Echo State Network performs reservoir-based feature transformation that captures dynamic communication patterns in the network, while the DTCCP classifier performs accurate classification of routing attacks based on the transformed features. The system also includes preprocessing techniques such as label encoding, missing value handling, and SMOTE-based data balancing to improve model performance. Experimental results demonstrate that traditional classifiers achieve moderate performance levels, while the proposed DRRD model significantly improves detection accuracy. Its high detection accuracy and scalable architecture make it suitable for real-world deployment in secure wireless sensor network environments