The IoT has emerged as a core enabling technology in a wide range of industrial applications and public service domains. Despite its widespread adoption, IoT infrastructures are inherently vulnerable to a variety of security threats that jeopardize data confidentiality, integrity, and service continuity. Furthermore, the heterogeneous nature of IoT-generated data, combined with noise, uncertainty, and rapidly changing operational conditions, significantly increases the complexity of detecting anomalous activities and compromised devices compared with conventional information technology networks. Consequently, designing accurate and dependable anomaly detection frameworks is critical to preventing malicious data from influencing IoT-driven decision-making processes. Although extensive research has been conducted in this area, many existing approaches still suffer from suboptimal performance. This limitation is often attributed to challenges such as class imbalance in security datasets, insufficient alignment between learning models and the underlying characteristics of IoT data streams, and suboptimal configuration of deep neural network parameters. To address these limitations, this study introduces an enhanced attack detection framework for IoT networks built upon a deep recurrent neural architecture, namely LSTM network. The proposed framework comprises three principal stages: baseline model training, metaheuristic-based parameter optimization, and comprehensive performance assessment. Initially, an LSTM model is trained on labeled training data, while a validation set is utilized for preliminary tuning and model selection. Subsequently, a PSO strategy is employed to further refine the network parameters, including connection weights and bias terms, with the objective of improving convergence behavior and detection capability. Finally, the optimized model is evaluated on an independent test dataset to examine its effectiveness in identifying malicious traffic patterns in IoT environments. Experimental demonstrate that the proposed method delivers superior detection performance compared to conventional deep learning-based methods, thereby providing a robust and efficient security mechanism for safeguarding IoT infrastructures against cyber threats.
Jalil et al. (Sat,) studied this question.