Fine‐Grained Hybrid Attention‐Aware Deep Learning Framework for Automated Multi‐Layer Cyberattack Detection in IoT‐Cloud Environments

The integration of Internet of Things (IoT) with cloud platforms has generated an extremely decentralized, data‐driven network that is transforming industries. At the same time, the integration poses intricate cybersecurity challenges because of scalability requirements, device heterogeneity, and an evolving threat environment. Conventional security metrics are inadequate in some circumstances, creating a necessity for adaptive, intelligent, and proactive defense modules. Artificial intelligence (AI) offers beneficial capabilities for intrusion detection, predictive threat modeling, anomaly detection, and malware analysis. Moreover, deep learning (DL) models have proved exceptional effectiveness in automatically learning intricate patterns from larger‐scale network traffic information, allowing precise identification of advanced cyberattacks in IoT‐cloud systems. Therefore, this manuscript presents a Fine‐Grained Hybrid Attention‐Aware Deep Learning Framework for Multi‐Layer Cyberattack Detection (FGHADL‐MLCD). The major intention of the proposed model is to enhance the reliability and intelligence of cyberattack detection in large‐scale IoT‐cloud infrastructures. The FGHADL‐MLCD approach operates through a sequence of stages, including preprocessing, attribute selection, cyberattack classification, and model optimization. Initially, the collected network traffic data from TON‐IoT and Edge‐IIoT datasets undergo preprocessing operations such as data cleaning, label encoding, feature scaling, and dataset splitting to ensure consistency of input data. Following that, an effective attribute selection mechanism is introduced using the Boruta feature selection technique, which identifies the most relevant network traffic features. For multi‐layer cyberattack detection and classification, the proposed model employs a bidirectional temporal convolutional network integrated with an attention mechanism. Furthermore, the training process of the deep learning model is optimized using the Adam optimizer. To assess the model's effectiveness, extensive experiments are conducted using the benchmark IoT cybersecurity datasets. The experimental results indicate that the proposed technique achieves improved cyberattack detection performance compared to recent models. Thus, the proposed model delivers a reliable solution for securing IoT‐cloud infrastructures against diverse and evolving cyber threats.