A New Study of an IoT Intrusion Detection and Membership Inference Attack
To adapt to the weak computing power of the Internet of Things nodes,insufficient storage space and the vulnerability of sensi-tive data attacks,a new lightweight intrusion detection model incorporating a convolutional neural network and differential privacy are proposed to make the model better adapted to the demanding resource environment of IoT.Firstly,the raw traffic data is pre-processed using the MinMax algorithm for normalization.Secondly,a lightweight convolutional neural network is designed to extract traffic features and perform classification.Finally,the differential privacy algorithm is used to defend against membership inference attacks that the model may encounter.The new algorithm is experimented on intrusion detection datasets such as UNSW_NB15,and the model accuracy reachs 98.98％,the precision rate reachs 98.05％,and the model size is controlled at about 200 KB,which improvs the accuracy rate by 2.81％compared with the DAE-OCSVM algorithm,and is suitable for the high accuracy intrusion detection required in the harsh envi-ronment of IoT.Meanwhile,the membership inference attack that the model may encounter are investigated,and the new algorithm re-duces 20.96％of the membership inference attack after incorporating the differential privacy algorithm.
intrusion detection systeminternet of thingsconvolutional neural networkdifferential privacymembership inference attack
国家科技期刊平台
NSTL
万方数据
