SDCBM: A Secure Data Collection Model With Blockchain and Machine Learning Integration for Wireless Sensor Networks

Wireless sensor networks (WSNs) often struggle with managing extensive data volumes, given their resource-constrained nature. Deployed in unattended areas, they face significant security risks and attacks. This study introduces the secure data collection model with blockchain and machine learning integration for WSNs (SDCBM), designed to identify intrusions and ensure secure data collection and storage for WSN applications. SDCBM employs an extreme learning machine (ELM) model, a fast single-layer feedforward neural network (NN), and integrates techniques for balancing the data distribution and selecting relevant features to enhance real-time detection of malicious attacks. Data is preprocessed and balanced utilizing the synthetic minority oversampling technique (SMOTE) and Tomek-Links combination method. To enhance the feature selection process, the Harris Hawk optimization (HHO)-based method is proposed. The blockchain module manages network node registration, authentication, node revocation, and secure storage of data hashes and node credentials. Simulation results demonstrate the efficacy of the proposed SDCBM method in detecting malicious nodes and enhancing secure data collection, thereby strengthening the security of WSNs.

I. Introduction

Wireless sensor networks (WSNs), comprising a group of spatially dispersed sensor nodes, are crucial in a variety of industries considering that they provide cost-effective, scalable, and adaptable solutions [1], [2], [3]. However, the distinct characteristics of WSNs, such as limited computing power, bandwidth, and vulnerability in hostile environments, make robust security measures essential to ensure their successful operation [4], [5]. Especially, denial-of-service (DoS) attacks are a persistent threat that demands effective approaches for detection and mitigation [6], [7]. While traditional network security has been considerably investigated, WSNs demand specialized and adaptive solutions, specifically due to the increasing data volume and relevance [8]. However, implementing an intrusion detection system (IDS) in WSNs raises further challenges compared to other systems due to their resource constraints [11]. Considering these challenges, gaining high detection accuracy while minimizing computational overhead is a high priority for WSN security [13].