Enhanced attack detection in IoT using multi-kernel support vector attention autoencoders

This study presents a deep learning-based technique for attack recognition in IoT networks, specifically using a hybrid model, named multi-kernel support vector attention autoencoder (MKSV-Attention Autoencoder), to detect botnet attacks. At first, traffic data is collected and then preprocessed by employing a number of steps to improve data balance and quality. Here, the missing data imputation, min–max normalization, and synthetic minority oversampling technique are exploited for preprocessing. Then, feature selection is done by using the LASSO model. Following that, attack detection in IoT is performed by utilizing the proposed MKSV-Attention Autoencoder. Finally, the experimentation evaluation is performed by employing metrics, such as accuracy, recall, precision, F1 score, TNR, and FAR. The assessment indicates that the proposed model attained a maximal accuracy of 0.9651, a precision of 0.9649, a recall of 0.9653, an F-measure of 0.9651, a TNR of 0.9649, a FAR of 0.0351, and a balanced accuracy of 0.9651.