Researchers have designed a distributed cyber defense architecture combining convolutional neural networks, long short-term memory networks, and blockchain technology across fog and cloud layers to detect cyber-attacks. Developed by Ali Kooshari, Mehdi Fartash, and Sara Nazari, the system deploys an intrusion detection mechanism that analyzes spatial and temporal traffic features while coordinating defense across decentralized network nodes. Testing across benchmark network environments demonstrated overall detection accuracies ranging from 98.85 percent to 99.87 percent.
The development addresses several fundamental operational hurdles in modern network security monitoring. Deep learning frameworks can identify complex attack signatures to some extent, but intrusion detection systems frequently encounter severe performance bottlenecks. In particular, the researchers highlighted that dataset imbalances, the high dimensionality of incoming network traffic, and the inherent complexity of model training present ongoing difficulties for standard intrusion detection systems operating in complex distributed environments.
To resolve data distribution challenges, the proposed framework implements data balancing directly within the fog layer before core classification occurs. The architecture balances incoming network traffic by integrating game theory with generative adversarial networks, while also injecting new attack methods into the training pipeline to bolster the system’s ability to recognize sophisticated, multifaceted cyber threats. Feature selection across the balanced data is subsequently executed by combining three distinct meta-heuristic methods to isolate optimal parameters.
The architecture divides computational tasks strategically between the fog environment and centralized cloud infrastructure. Within the cloud layer, a neural network combines convolutional neural networks and long short-term memory models to extract and evaluate final spatial and temporal traffic features. The combined model is trained with an optimized support vector machine output layer, establishing an integrated classifier that couples deep feature extraction with support vector discrimination before distributing the model parameters back to edge infrastructure.
Once trained in the cloud, the hybrid classifier is transmitted downward to active fog nodes across the network. Each fog node applies the optimized feature vector to reduce the model structure, allowing local nodes to efficiently detect anomalous attack traffic in real time. To maintain network-wide awareness, fog nodes exchange lists of identified attacker network addresses using a blockchain-based messaging framework, which preserves data confidentiality while coordinating defensive intelligence among distinct intrusion detection systems.
Empirical evaluations across multiple benchmark security datasets revealed high detection efficacy under varied attack conditions. On the NSL-KDD dataset, the framework achieved 99.65 percent sensitivity, 99.37 percent precision, and 98.86 percent overall accuracy. In evaluations using the UNSW-NB15 dataset, the architecture recorded 98.85 percent accuracy, 98.67 percent sensitivity, and 96.52 percent precision. Testing on the Edge-IIoTset dataset yielded 99.87 percent accuracy, alongside 98.79 percent sensitivity and 98.44 percent precision.
The comparative findings highlighted measurable improvements over existing data-balancing techniques and baseline learning models. On the NSL-KDD dataset, the framework exceeded the performance of standard balancing methods, including synthetic minority over-sampling, generative adversarial networks, and variational autoencoder generative adversarial networks. Against a base generative adversarial network model, accuracy improved by 1.49 percent on NSL-KDD, 3.27 percent on UNSW-NB15, and 3.2 percent on Edge-IIoTset. On Edge-IIoTset, the design also surpassed standalone convolutional neural networks, long short-term memory networks, and baseline combined models by 1.02 percent, 0.16 percent, and 0.36 percent, while broader tests across NSL-KDD, UNSW-NB15, Bot IoT, and Edge IoT confirmed higher detection accuracy than recurrent neural networks and gated recurrent units.
Subject of Research: Distributed intrusion detection using deep learning and blockchain across fog and cloud layers
Article Title: Blockchain-based distributed cyber defense in fog and cloud layers using CNN and LSTM
Article References: Kooshari, A., Fartash, M., & Nazari, S. (2027). Blockchain-based distributed cyber defense in fog and cloud layers using CNN and LSTM. Cluster Computing, 30(1), Article 17. https://doi.org/10.1007/s10586-026-06642-4
Image Credits: AI Generated
DOI: 10.1007/s10586-026-06642-4
Keywords: Intrusion Detection Systems, Cyber Defense, Fog Computing, Cloud Computing, Convolutional Neural Networks, Long Short-Term Memory, Support Vector Machines, Blockchain Messaging, Game Theory, Generative Adversarial Networks, NSL-KDD Dataset, UNSW-NB15 Dataset, Edge-IIoTset Dataset, Bot IoT Dataset, Feature Selection, Network Traffic Balancing, Data Confidentiality, Meta-Heuristic Methods
News Source: Blake Davidson. (November 16, 2026). Hybrid Deep Learning and Blockchain System Improves Distributed Cyber Attack Detection. Scienmag.



