A dimension reduction model and classifier for anomaly-based intrusion detection in Internet of Things

Internet of Things (IoT) devices and services have gained wide spread growth in many commercial and mission critical applications. The devices and services suffer from intrusions, attacks and malicious activities. To protect valuable data transmitted through IoT networks and users’ privacy, intrusion detection systems (IDS) should be developed to match with the characteristics of IoT, which requires real-time monitoring. This paper proposes a novel model for intrusion detection which is based on dimension reduction algorithm and a classifier, which can be used as an online machine learning algorithm. The proposed model uses Principal Component Analysis (PCA) to reduce dimensions of dataset from a large number of features to a small number. To develop a classifier, softmax regression and k-nearest neighbour algorithms are applied and compared. Experimental results using KDD Cup 99 Data Set show that our proposed model performs optimally in labelling benign behaviours and identifying malicious behaviours. The computing complexity and time performance approve that the model can be used to detect intrusions in IoT.