| （英） |
With the proliferation of IoT devices, IoT security problems arise. To protect heterogeneous connected devices in IoT networks against cyber-attacks and various attack patterns by intruders, many researchers have introduced network intrusion detection systems (NIDSs) which are based on machine learning techniques. An NIDS in IoT networks must maintain the appropriate security level despite the limited computational resources. To address the limitation, we propose a classification method for detecting the attacks by intruders to realize the NIDS designed for IoT networks. Through numerical experiments using a realistic botnet dataset in IoT networks with imbalanced class distribution, we demonstrate that the proposed classification yields high area under the receiver operating characteristics curve (AUC) score as well as balances the high accuracy with low false-positive rate, with the help of the synthetic minority over-sampling technique (SMOTE). |