Research On Intrusion Detection Methods For Multimedia Internet Of Things Based On Machine Learning

The evolution of Io T,its integration,heterogeneous protocols and deployment of numerous multimedia physical objects such as smartphones,smart refrigerators,security cameras etc.has created many new security challenges.An instance is,Imperva – a security firm,on March and April 2019,reported massive Botnet attack that used more than 400,000 Io T devices.Also,in 2019,security researchers from FSecure issued a report claiming 300% surge on cyberattacks on Io T devices.With this significant increase in multimedia big data generated from Io T devices,the harvested digital information become susceptible to diverse network attacks hence the need for a more secured technology to combat these attacks.In order to implement a more efficient method to maintain data confidentiality,integrity and also ensuring the availability of resources when needed,this thesis researches on machine learning-based intrusion detection algorithms to deal with drawbacks such as low detection accuracy,data redundancy issues,high false positive rate and low performance rate some existing conventional IDSs pose to information security in Multimedia Internet of Things(MIo T).The main research work in this thesis is presented as follows:1.To improve on the low detection accuracy and the high false positive rate in most existing IDS techniques to further secure digital information harvested from the numerous Io T devices,an efficient machine learning IDS based on Principal Component Analysis(PCA)and Support Vector Machine(SVM)is proposed in this thesis.Firstly,data dimensionality's feature extraction is performed to extract the relevant and important features in the NSL-KDD dataset by adopting PCA to reduce the computational load and also enhance the detection efficiency of the intrusion detection model.Secondly,the pre-processed data is used to train our model using the SVM classifier while tuning its parameters in order to achieve a good classification accuracy.Finally,our trained model is tested on the test dataset to be classified into either normal or attack.Experimental results show that the proposed method attained a good classification accuracy with a low false positive rate value when compared with different classifiers and other existing techniques.2.In order to improve the classification accuracy and enhance the exponentially increasing noisy nature of network traffic data produced by the diverse types of Io T devices,an effective IDS based on Information Gain(IG)and C4.5(J48)decision tree algorithm is proposed in this thesis.Firstly,the proposed method adopts IG feature selection method to select the relevant and important features from the KDD'99 and NSL-KDD datasets.Afterwards,the pre-processed network data is trained using the C4.5 decision tree algorithm while tuning its parameters so to achieve a good classification value.Finally,classification is performed using the test dataset with our trained model.In testing our system,experimental results show that the proposed approach can detect malicious attacks with high detection rate,low false positive rate at a high training speed.