Learning-based Intrusion Detection Methods And Their Application To IoT Security

Internet of Things(IoT)has been developing at full speed and gradually scaling up,which leads IoT security to gain more attention and face bigger challenges.Intrusion is one of the most significant issues in IoT network security.Since the software systems have become to be various,research on intrusion detection system(IDS)with high flexibility and portability is essential.This thesis mainly focuses on machine learning based intrusion detection methods and their applications to IoT security,and the feature processing's impact to intrusion detection is also discussed in this thesis.The proposed algorithms are tested on intrusion detection benchmark NSL-KDD dataset,and the corresponding detection process comprises preprocessing,scaling and classification.The numeralization of features in the NSL-KDD dataset is adopted in the preprocessing.To solve the problem of the unsteady feature ranges between the training dataset and testing dataset,nonlinear scaling schemes have been proposed.The support vector machine(SVM)method with kernel function is used for the classification.On the scaling stage,both nonlinear and linear scaling methods have been tested on the NSL-KDD dataset.Experimental results show that compared to the linear scaling method,the nonlinear scaling method can achieve better performances by improving the total accuracy of binary-classification test from 79% to 82.2%,and accuracy of DoS(Denial of Service)detection in multi-classification test from 76.7% to 86.5%.Moreover,this thesis analyzes DoS attack types in detail,and simplifies the features in intrusion detection process.On one hand,the NSL-KDD dataset has been reprocessed and the SVM based classification model has been retrained.On the other hand,DoS attacks have been implemented towards the IoT platform to attain the connection features,which would go through preprocessing and scaling stages and then judged by the retrained model.The experimental results demonstrate that the applicability of SVM-based intrusion detection algorithms in IoT security,and reveal that the appropriate feature scaling methods have greatly direct impact to the detection effects.