DeepSVM-A Novel Approach for Early Detection and Classification of IoT Botnet Attacks
Veena Antony, Nainan Thangarasu · 2024
Though there are few security mechanisms developed to detect the IoT botnet attack, they are mostly rule based classifiers, which are formal rule-based detection that could be circumvented by the malware attacker’s knowledge. This research study deals with the problem of overcoming the issue of determining patterns from the voluminous IoT dataset and predicting the attacks and normal packets in presence of class imbalance. To enhance the accuracy rate of IoT botnet attack detection by developing a deep learning paradigm to discover the pattern of malicious packets before they penetrate the host network. To address the overfitting issue that arises during the training phase of a deep neural network, the proposed model called Deep Neural Support Vector empowered with Butterfly Optimization Algorithm (Deep-SVM+BOA) replaces Support Vectors to identify the pattern of incoming data and categorize it as malicious or benign packets. To determine if a message is anomalous or not, it makes use of the attention mechanism and the fully connected layer network. Instead of using SoftMax, Support vector machine is used in this proposed work to perform classification and the hyperparameters values are scrutinized by the food searching behavior of the Butterfly Optimization Algorithm. The simulation results proved the efficiency of the proposed Deep-SVM+BOA compared with other three models using three different evaluation criteria to detect the IoT botnet attack on a UNSW-NB15 dataset.