Security challenges due to the large-scale, heterogeneous, and distributed nature of IoT deployments

The gap

Evidence profile

Sourced from the recommendations and future work and stated challenges and stated research gap of the source papers, classified as general, drawn from work published between 2024 and 2026, spanning 4 journals. Those papers have been cited 126 times in total.

Supporting evidence — 5 representative gaps

• A survey on intrusion detection system in IoT networks (2024) · Cyber Security and Applications · cited 89× · doi

  Exploring alternative deep learning models and optimization techniques to further enhance performance and reduce computational overhead. developing more adaptive and real- time detection mechanisms for DDoS attacks in IoT environments. exploring further optimization strategies to reduce the computational complexity of the model, enabling its deployment in resource-constrained environments.

  generalrecommendationsevidence 5/5
  Keywords: exploring optimization further reduce computational environments alternative deep learning models techniques enhance performance overhead developing
• SLArDHO-Net: adaptation multi-model detection of IoT malware with self-learning dynamic hyperparameter optimization (2026) · Discover Applied Sciences · doi

  ARTICLE IN PRESS ARTICLE IN PRESS ACCEPTED MANUSCRIPT ARTICLE IN PRESS The suggested SLArDHO-Net model proves high levels of performance in semi-supervised detection of internet malware, with macro-F1 scores of 0.97-0.98 on CIC-IoT-2023 and stable values of 0.90-0.96 on Edge-IIoTset and IoT-23 data. Smart rebalancing with the help of dynamic hyperparameter optimization using SLArDHO, hybrid temporal and static classification, and threshold models is effective in addressing the main issues such as limited labelled data, extreme imbalance of the classes, and computational requirements under dynamic IoT conditions. Such contributions allow efficient intrusion detection operation even with limited supervision, which is useful in the practical deployment against the zero-day and dynamic threats. Despite such strengths, it is also susceptible to pseudo-label quality in near zero labeling regimes where noise propagation can exist by using erroneous initial clustering, and simulated edge metrics are used in place of real-world deployments of hardware. Future directions will focus on federated learning to train with privacy guarantees across the nodes of an IoT and SLArDHO will be expanded with reinforcement learning agents to facilitate proactive detection of threats and automated threat mitigation measures. To improve the privacy and scalability of the framework, we will strive to incorporate federated learning with IoT devices so that local models can be trained on the global models without requiring locally sensitive data to be shared in the future. This decentralized system will also enhance data diversity and strength, as well as the maintenance of user privacy. Furthermore, we will apply the SLArDHO algorithm to reinforce learning-based agents that will allow us to conduct threat hunting and mitigation proactively. Such agents will be able to understand ideal security policy dynamically, guided by environment response and attack indications, and transform their direct detection capabilities to proactive, self-directed defense systems.

  generalfuture workevidence 5/5
  Keywords: slardho detection learning article press dynamic models privacy agents edge using limited allow zero threats
• An Efficient Flow-Based Anomaly Detection System for Enhanced Security in IoT Networks (2024) · Sensors · cited 29× · doi

  This paper presents several important contributions to the field of IoT security. The primary contribution is the introduction of the IoT-FIDS, a novel profiling algorithm specif- ically designed for intrusion detection in IoT environments. Unlike traditional machine- learning-based methods, the IoT-FIDS leverages flow-based packet representations to detect abnormal packets and flows, providing a highly efficient and lightweight alternative. By avoiding the computational complexity associated with machine learning, the IoT-FIDS is particularly well suited for resource-constrained IoT devices, making it a more practical so- lution for real-time deployment. A second key contribution is the development of an anomaly-based detection mecha- nism that operates without the need for machine learning models. The IoT-FIDS signifi- cantly reduces the overhead typically involved in training and updating models, offering a more scalable solution that can be deployed at both the network and host levels. This effi- ciency, combined with its minimal resource consumption, positions the IoT-FIDS as a supe- rior alternative to machine-learning-based methods, particularly in environments where computational power is limited. Furthermore, the efficacy and performance of IoT-FIDS were rigorously validated through comprehensive testing using two well-established IoT datasets, UNSW-NB15 and BoT-IoT. These evaluations demonstrate that the IoT-FIDS not only achieves high detection accuracy but also maintains a low false-positive rate, outperforming many machine-learn- ing-based intrusion detection systems in terms of both precision and resource efficiency. Sensors 2024, 24, 7408 21 of 23 This robust performance underlines the practicality of IoT-FIDS for real-world IoT deploy- ments, where low latency and efficient resource usage are critical. However, one limitation of this study is that the complexity analysis of advanced opti- mization techniques, such as Bayesian Hyperparameter Optimization, was not included. Future work should address this by analyzing the complexity and performance impact if such optimization methods are applied to deep neural networks (DNNs) for intrusion de- tection in IoT environments. This could provide valuable insights into optimizing detection models while balancing resource constraints. Another limitation of this study is that the pro- posed IoT-FIDS does not specifically address MQTT-based IoT communication systems, which use TCP and SSL encryption for secure transmission. Given the widespread adoption of MQTT in IoT networks, future work should focus on optimizing the IoT-FIDS to handle the smaller packet headers and encrypted traffic typical in such environments, while ensur- ing effective intrusion detection. There are still several avenues for improvement. As a network-based system, the IoT- FIDS heavily relies on its ability to analyze suff

  generalfuture workevidence 5/5
  Keywords: fids based detection machine resource intrusion environments learning complexity models performance several contribution packet efficient
• An Energy-Aware Security Framework for the Internet of Things Integrating Blockchain and Edge Intelligence (2026) · Computers · cited 1× · doi

  Security challenges due to the large-scale, heterogeneous, and distributed nature of IoT deployments - Scalability challenges due to the limited computational capability, communication capacity, and power resources of IoT devices - Energy efficiency challenges due to the need to reduce energy consumption while maintaining strong cybersecurity protection and real-time anomaly detection

  generalstated challengesevidence 5/5
  Keywords: security challenges due large-scale heterogeneous distributed nature iot
• Hybrid GNN–LSTM Architecture for Probabilistic IoT Botnet Detection with Calibrated Risk Assessment (2026) · Computers · cited 7× · doi

  The paper identifies a gap in existing intrusion detection systems, which treat network events as independent observations, ignoring important patterns. The gap is the lack of consideration for device relationships and temporal phases in IoT botnet detection. The paper also notes that most devices were designed with functionality as the priority, not protection, resulting in an enormous attack surface.

  generalstated research gapevidence 5/5
  Keywords: paper identifies gap existing intrusion detection systems treat

Questions about this gap