Reliable communication and accurate data collection are crucial tasks in Wireless Sensor Networks (WSNs). Due to the lack of having no central communication infrastructure, WSNs can be exposed to various attacks. One of the common attack types in WSNs is Byzantine attack, in which the attacker can reduce the reliability of the network by adding new nodes to the network area and sending fake data. This study proposes two ensemble-based approaches for detecting the Byzantine attacks in WSNs. The proposed approaches combine three different traditional classification algorithms (Naive Bayes, decision tree (C4.5), and k-NN) with voting and stacking methods. In addition to the proposed methods, the current ensemble learning approaches (C4.5 based Bagging (Bagging(C4.5)) and Boosting (AdaBoost)) and the traditional algorithms (Naive Bayes, C4.5 and k-NN) were applied on a sample network of 66 IRIS nodes (60 normal, 6 malicious) within experimental studies. The classification results obtained from each algorithm were compared according to the accuracy rate and f-measure values. The results gathered from the testbed show that the ensemble-based methods achieve up to 98.48% accuracy rate for detection of the Byzantine attacks in the sample network while this ratio for the traditional methods is limited to the 96.97%. In large networks with more nodes, the difference among these ratios may increase.
Alan : Mühendislik; Fen Bilimleri ve Matematik
Dergi Türü : Ulusal
Benzer Makaleler | Yazar | # |
---|
Makale | Yazar | # |
---|