Detection of Denial-of-Service Attack in Wireless Sensor Networks: A lightweight Machine Learning Approach

• Published in: IEEE access Vol. 11; p. 1
• Main Author: Elsadig, Muawia A.
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.01.2023; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Classifiers; Computer security; cyber security; Decision trees; Deep learning; Denial of service attacks; Denial-of-service attack; DoS attacks; Feature extraction; feature selection; Internet of Things; IoT security; Lightweight; Machine learning; Network security; Routing; Security; Wireless sensor networks; WSN attacks; WSN constraints; WSN security
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2023.3303113

The characteristics and performance of wireless sensor networks (WSNs) are the main reasons for their rapid expansion in various fields. However, these networks are extremely susceptible to multiple security assaults, including denial-of-service (DoS) attacks, which are among the most prevalent in these networks. This study sheds light on WSN restrictions, weaknesses, and security threats with a focus on DoS attacks. Recent techniques for DoS attack detection have been investigated thoroughly, highlighting their achievements and limitations. This provides valuable insight into the current state of recent research in this field. Accordingly, this study proposes a lightweight machine learning detection approach based on a decision tree (DT) algorithm with the Gini feature selection method to detect DoS attacks in WSNs. An enhanced version of the WSN-DS dataset, developed by the author, was used to train and test the proposed approach. The proposed approach has shown good performance by achieving an accuracy rate of 99.5% with minimum overhead compared to random forest (RF), extreme gradient boosting (XGBoost), and k-nearest neighbor (KNN) classifiers. It only takes 9.7%, 13%, and 2% of the processing time required by FR, XGBoost, and KNN respectively, which indicates that our proposed approach significantly outperforms these classifiers in terms of processing time. It is noteworthy that RF achieved an accuracy that was somewhat superior; however, the proposed approach greatly surpassed RF by taking only 9.7% of the RF processing time, which is an important factor in meeting WSN constraints.