Fair Consensus in Blockchain‐Dual Sampling Dilated ConNet: An Optimized Intrusion Detection and Prevention System in IoT

ABSTRACT The rapid IoT device proliferation has greatly increased the attack exposure, making IoT networks highly susceptible to cyber threats. Conventional intrusion detection systems (IDSs) have trouble with the complexity of IoT networks due to their massive, heterogeneous, and real‐time data streams. It is the immense amount of information produced by various IoT devices (e.g., sensors, cameras, and wearables) that send data streams continuously in real time. This impacts IDSs as it becomes difficult to process, analyze, and identify threats rapidly and precisely. The variety and velocity of data can overwhelm conventional IDSs, resulting in delays, ignored attacks, or false positives, particularly under limited computational resources common in IoT settings. Existing IDS methods frequently have poor scalability, large false‐positive rates, and insufficient real‐time threat detection, failing to ensure both data security and privacy. To address these issues, this paper introduces an optimized deep learning‐driven model called dual sampling dilated pre‐activation residual attention convolutional neural network optimized using greylag goose optimization (DSD‐PRA‐ConNet‐GGO), which integrates blockchain technology for intrusion detection and prevention in IoT networks. The methodology starts with data acquisition from various IoT devices and sensors, such as smart cameras and environmental sensors, to capture traffic patterns and potential intrusions. The system shows superior performance, achieving 15.59% to 36.88% higher accuracy compared to existing methods such as LSTM‐IDS, ML‐XGBoost, RNN‐IDS, and DL‐DFCN, based on evaluations conducted using the BoT‐IoT dataset, which includes a wide range of realistic attack scenarios and benign traffic commonly encountered in IoT environments.