[en] In recent infrastructures, Internet of Things (IoT) have become an important technology for connecting various actuators and sensors over wireless networks. Due to increase in mission-critical infrastructures, we make use of these new technologies for reliable communication but their security is always not promising in terms of availability, confidentiality, integrity, and privacy of network services. Users can be compromised and vulnerable by a motivated malicious opponent unless they are not adequately protected by a robust defense. Due to this reason, an ambient intelligence approach for Intrusion Detection System (IDS) is required. In this research, ww proposed Ambient Approach based on Reinforcement Learning Integrated Deep Q-Neural Network (RL-DQN) model for WSNs and IoT in which it leverages the Markov decision process (MDP) formalism to enhance the decision performance in IDS. We deploy RL-DQN-IDS over Edge-cloud intrusion detection infrastructure in which binary attack classification of the network traffic is performed at the edge network while multi-attack classification is performed at the cloud network. To identify intrusions, we use a two-phase process that includes an initial learning phase that relies on RL, followed by a detection and classification phase that relies on DQN. We used four datasets namely UNSW-NB-15, BoTNeTIoT-L01, CICIDS2017 and IoTID20 with a smart house simulation environment configured with WSN and IoT technologies to evaluate performance. Accuracy, precision, and recall were all considered while assessing the dataset under consideration. When compared to five other machine learning models, the RL-DQN model method has demonstrated superior performance. This model outperforms the other five that were tested.
[University of Luxembourg > Faculty of Science, Technology and Medicine (FSTM) > Department of Mathematics (DMATH) >]
