Akademik Veri Yönetim Sistemi
IEEE INTERNET OF THINGS JOURNAL, cilt.11, sa.2, ss.2109-2125, 2024 (SCI-Expanded)
The popularity of wireless sensor networks (WSNs) increases as the usage areas and the number of integrated systems increase, and this situation attracts the attention of attackers. Attackers carry out attacks aimed at infiltrating, capturing, and manipulating the network. These attacks are implemented differently according to the layers. After these attacks on sensor networks, network traffic data is examined and malicious traffic and node behaviors are analyzed to prevent possible future attacks. The raw data received from the network are made usable by learning models by some preprocessing. The data analyzed with the models are categorized according to the network traffic types and the attacks carried out on the network are detected. In WSN, attack detections made with learning models are performed with high-accuracy percentages compared to classical detection methods. In this study, blackhole, flooding, and selective forwarding attacks, which are WSN network layer attacks, were created and implemented in network scenario (NS 2) simulation environment. The WSN-BFSF data set was obtained, and the obtained data set was made ready to be examined with learning models after the necessary preprocessing. The WSN-BFSF data set consists of 312 106 rows. The data set was examined with four different machine learning models random forest, decision tree, Naive Bayes, and logistic regression, and eight different deep learning models multilayer perception (MLP), convolutional neural network (CNN), long short-term memory (LSTM), gated recurrent unit (GRU), CNN-LSTM, LSTM-CNN, CNN-GRU, and GRU-CNN. The experimental results obtained with the models are presented in detail.