Internet of Things (IoT) devices are increasingly susceptible to botnet threats, with Mirai-based attacks posing significant security challenges. To address these concerns, we present a lightweight anomaly detection model based on a Gaussian Mixture Model (GMM), tailored for detecting Mirai botnet traffic. By leveraging anomaly detection techniques, our approach identifies malicious activity with optimized latency and computational efficiency, making it suitable for resource-constrained IoT environments. Trained specifically on Mirai-related traffic, the proposed model achieves comparable detection accuracy while significantly improving latency compared to traditional decision tree-based models. These results underscore the effectiveness of the GMM approach in balancing detection performance and real-time responsiveness for IoT applications.

International audience

5G networks are highly dynamic and non-homogeneous networks, making resource management more complex and vulnerable to different network attacks like DDoS, Port Scanning, etc. In addition, Network Slicing plays an important role in these networks in enabling a multitude of 5G applications, services, and use cases. In this context, we propose in this paper, a new approach to secure 5G network slices by developing a models' orchestrator providing an adaptive Machine-Learning (ML) models as-a-Service. Specifically, the proposed models' orchestrator is a cloud server that acts as a decision-making entity to offer on-demand adaptive ML models to detect potential attacks by tuning and adapting ML parameters and algorithms according to the characteristics of the requester devices/nodes and the real-time conditions of each network slice. We demonstrate the effectiveness of our approach through a series of experiments, by training different ML algorithms with different network slices properties. Results show that our approach provides a more efficient and effective way of securing 5G networks compared to traditional methods in terms of respecting the requirements raised by each slice/node of these networks.

The Internet of Things (IoT) is starting to integrate deeply into our daily lives thanks to the different services it provides. This technology has already made us more closely linked to the external environment through ubiquitous communication devices. However, even though this proximity has numerous benefits, it also has a significant security impact, where the cyber-attack surface has grown dramatically. In this regard, we present, in this paper, our results toward the development of a decision tree-based machine learning model for intrusion detection in Massive-IoT networks. The principal objective of this work is to provide a highly accurate detection model, while preserving resource consumption by developing a real prototype of the intrusion detection system. To this end, we first propose and apply our pre-processing methodology on the well-known Avast IoT-23 dataset, allowing us to reach a high detection rate with 99.99% of accuracy and just 1804KB of the model’s size. Then, we propose a new machine learning model based on the decision tree classifier and deploy it in a real environment with malicious attack traffic. Obtained results show that our proposed model allows 88% of real-traffic-based precision rate and up to 90% of specificity.

Internet of Things starts to integrate deeply our daily life through the comfort and the services that it offers. This technology made us already more intertwined with the external environment via deployed communicating devices. However, even though this proximity offers many advantages, it presents also strong security issues. The cyber-attack surface is significantly increased and the intruder impact is becoming more compromising. In this context, we present the preliminary results of our study toward the conception of a machine learning based solution against cyber-intrusions. The main purpose is to develop a resource-preserving solution with high precision of detection. We were interested in the IoTID20 dataset over which we experimented the most accurate machine learning models. The employed pre-processing approach that we propose pushes the machine learning models to the peak of their performances. Our study provides a comprehensive view of models efficiency with respect to detection rate, size and delay.

Intelligent traffic light control systems are proposed to solve the problem of congestion in urban areas. In this context, the connected vehicles are used to gather the real-time traffic data surrounding environment. The involvement of vehicles in the control of traffic can reduce considerably the travel delay and maximize throughput. However, if the reported information is fake, then the result could be catastrophic, such as road accidents. Also, the malicious vehicles can send fake displacement urgency to their supervised traffic light, to reach their destinations as fast as possible. To tackle this problem, we propose a Reputation model based Intelligent COntrol Protocol (RICOP). In this protocol, each traffic light computes the negotiating vehicle's reputation based on its dynamic behavior, the opinions of located traffic lights, and its historical reputation monitored by nearest roadside unit (RSU). According to the reputation value, the traffic light takes an appropriate decision. RICOP guarantees the integrity and authentication of exchanged messages using digital signature and certificateless cryptography. Also, the privacy of the vehicle and its driver is preserved using a pseudonym. To evaluate the effectiveness of our proposal, we set a series of simulations in a network simulator NS3. The results are satisfactory in terms of delay, accuracy and computational cost.

The emergence of IoT technology can integrate connected digital wearable devices in the health field. In this context, the smart hospital is created to facilitate and improve the quality of patient's medical life. However, uncontrolled access to personal and health information of patients can disrupt the smooth functioning of smart hospital services. Consequently, the protection of patient privacy remains a big issue that must be addressed. In this paper, we propose a Context-Aware Privacy Model (CAPM). It aims to secure the exchanged and shared patient's information during his hospital stay. To evaluate effectiveness of our proposal, a series of simulations are carried out using NS3. The obtained results demonstrate the efficiency of our protocol in terms of communication delay, Authentication delay, and pseudonym generation time.

The traffic problems related to bottlenecks, accidents and delays on the road intersections create a significant need for the application of platoon mechanism. In this context, we consider a scenario when a platoon of negotiating vehicles approaches the intersection. The platoon leader should negotiate the passage of its following vehicles from the traffic light to enable them to reach their destinations as soon as possible. However, some platooned member vehicles try to destroy platoon stability. In this paper, we considered the VPD (Vehicle platooning disruption) attacks, where an attacker seeks to destabilize or take control of a platoon through false data injection (FDI) and replay of control messages. To conceive a countermeasure to this attack, first, we propose a VPD Attacks Detection Algorithm (VPD-ADA), which can identify the instability of the platoon dynamics model as a primary step for mitigating the impact of FDI and replay attacks. We use stochastic time series analysis to evaluate the impact of VPD attacks by the deviation anomalies of cartesian coordinates of vehicles from the expected trajectory determined by the platoon. Second, we introduce a Reputation-based Reliable Mitigation Algorithm (RRMA) that uses fused information from neighboring vehicles to decide the reliability of received messages.

Vehicular networks have emerged as a promising technology for the development of traffic management systems in smart cities. They are expected to revolutionize a variety of applications such as traffic monitoring and pay-as-you-drive services. Recently, the notion of road pricing has become crucial in most big cities as it contributes in road congestion avoidance, fuel consumption saving, and pollution reduction. However, as the road pricing systems need trip data to invoice citizens, it is vital to ensure geolocation privacy while keeping drivers honest. In this paper, we propose a security approach for smart road pricing systems, which prevents toll evasion violations. The proposed approach operates under a fully distributed threshold-based control system to detect fraudulent drivers trying to cheat on their tolls. The accused drivers are reported to the toll server in order to take the appropriate countermeasures. Through the security analysis, we show the robustness of the proposed approach against a range of potential attacks. We also evaluate the proposed approach through simulations considering important metrics, namely, the storage and communication overheads. The proposed approach shows better performance results in comparison to the existing approaches. Furthermore, we evaluate the proposed approach efficiency in terms of detection precision, where it demonstrates promising results.

The proliferation of network technologies drives many different network architectures to provide huge variety of services and contents to end clients. This task becomes more difficult when we are in networks with intermittent connections, called delay tolerant networks (DTN) where security is an important issue. In this paper, we propose a trust-based approach to secure data transfer in DTN in the presence of malicious transporters. Our proposal is intended to a DTN architecture which includes several sub-networks geographically dispersed in isolated regions and having an intermittent access to an infrastructure-based network (like internet). Our approach is based on a particular web-of-trust, which is formed based on existing social relationship among clients and transporters. We conducted intensive simulations and the obtained results show that it offers high packet delivery rate and resists against malicious transporter's behaviour.

Wireless sensor network (WSN) is often deployed in hostile environments, which make it vulnerable to attacks. WSN comprises a large number of sensor nodes with different hardware abilities and functions. Due to the limited memory resources and energy constraints, complex security algorithms cannot be used in WSN. Hence, it is necessary to balance between security requirements and energy consumption. In this paper, we propose a lightweight public-key-based authentication protocol for WSNs. Our protocol uses identity-based encryption in order to lighten the energy consumption from the public-key certificate management. We have developed two variations of the proposed solution depending on the manner and when an attack should be detected. We perform an overall evaluation of our approach through simulations. The results indicate out performance of our approach in terms of energy consumption while providing effective security.