Botnet Attack Detection Using Convolutional Neural Networks in the IoT Environment
Kübra Nilgün Karaca, Aydın Çetin · 2021 International Conference on INnovations in Intelligent SysTems and Applications (INISTA) · 2021
In this study an experiment has been conducted on the detection of the most common botnet attacks in the Internet of Things (IoT) environment, using Convolutional Neural Networks (CNN) architecture. The developed CNN model was optimized by fluctuating the values of the hyperparameters at each repeating run. After the optimization process, the model was evaluated and obtained results were compared with the results in the literature. The entire dataset contains nine datasets of nine different IoT devices, each has ten types of botnet attacks. One of these datasets was used in the experiments. The proposed CNN model classified benign network traffic and botnet network traffic correctly, with a success rate (accuracy) of 97.98%. It performed better than Gaussian Naive Bayes (GNB), ANN, Support Vector Machines (SVM), and K-Nearest Neighbors (KNN) models which have success rates of 75.98%, 88.70%, 97.14%, and 97.48%, respectively.