A Novel Approach for Cyber-Attack Detection in IoT Networks with Neutrosophic Neural Networks

Abstract

The exponential expansion of Internet-of-Things (IoT) devices has created complex, 
interconnected ecosystems, exposing them to an increasing range of sophisticated cyber threats. 
Existing intrusion detection systems in IoT environments often fail to handle uncertainty and adapt 
to evolving cyber threats, leading to high false-positive rates and reduced reliability. To address these 
challenges, this study proposes a hybrid cyber-attack detection model that integrates neutrosophic 
set theory with deep neural networks. Neutrosophic method makes the uncertainty model more 
flexible since it sets the truth, indeterminacy, and falsity values to the network traffic data, and the 
neural network is used for adaptation and classification accuracy. The model is different from the 
traditional machine learning techniques that use crisp data presentations: our model brings a 
neutrosophic-approach-based uncertainty quantification system that significantly turns the machine 
model into a resilient one exposing to zero-day and adversarial attacks. From the experimental 
findings it is obvious that the proposed Neutrosophic Neural Network model not only could detect 
IoT cyber-attacks, but it was also much more accurate. The model got an accuracy of 95.8%, with the 
whole set of items set out in the table for 93.2% precision, 92.5% recall, whereas the previous F1-score 
turned out to be 92.8%. These are the outcomes of the investigation that prove that our neutrosophic 
enhanced AI models are the most effective tools for the escape from security issues that appear in IoT 
environments and that provide the full security.

DOI: 10.5281/zenodo.15540377