The Internet of Things (IoT) is vulnerable to various attacks, due to the presence of tiny computing devices. To enhance the security of the IoT, this paper builds a lightweight intrusion detection system (IDS) based on two machine learning techniques, namely, feature selection and feature classification. The feature selection was realized by the filter-based method, thanks to its relatively low computing cost. The feature classification algorithm for our system was identified through comparison between logistic regression (LR), naive Bayes (NB), decision tree (DT), random forest (RF), k-nearest neighbor (KNN), support vector machine (SVM) and multilayer perceptron (MLP). Finally, the DT algorithm was selected for our system, owing to its outstanding performance on several datasets. The research results provide a guide on choosing the optimal feature selection method for machine learning.

The main problem facing the Internet of Things (IoT) today is the identification of attacks due to the constrained nature of IoT devices. To address this problem, we present a lightweight intrusion detection system (IDS) which acts as a second line of defense allowing the reinforcement of the access control mechanism. The proposed method is based on a Deep Auto-Encoder (DAE), which learns the pattern of a normal process using only the features of the user’s normal behavior. Whatever deviation from the expected behavior is considered an anomaly. We validate our approach using two well-known network datasets, namely, the NSL-KDD and CIDDS-001. The experimental results demonstrate that our approach provides promising results in terms of accuracy, detection rate and false alarm rate.