A Zero-Shot Learning-Based Detection Model Against Zero-Day Attacks in IoT
Xueqin Gao, Kai Chen, Yufei Zhao, Peng Zhang, Longxi Han, Daojuan Zhang · 2024
With the large-scale deployment of loT devices, loT security faces huge threats from zero-day attacks, and there is an urgent need to conduct research on loT zero-day attack detection technologies. Aiming at the problem of lack of available samples for zero-day attack detection in loT, we propose a zero-day attack detection method based on generative zero-shot learning. Feature extraction is performed based on the LightGBM model, and then multi-layer perceptron is used to generate attribute vectors. We generate pseudo feature space of data category based on conditional variational autoencoders to train zero-day attack detectors. To comply with real attack detection scenarios, we adopt the generalized zero-shot learning method to conduct verification on the UNSW-NB15 data set. Experiment results show that the precision, recall and Fl of most attack types are above 90%. The average accuracy of our method in detecting zero-day attacks is 93.6%, which is much higher than the baseline method and achieve great detection results.