Absract

The Fuzzy Deep Neural Network for Cybersecurity in Edge/IoT systems (FDNN) is a novel deep neural network architecture developed by incorporating fuzzy numbers into traditional artificial neural network structures. This approach enables the development of new cybersecurity solutions for anomaly detection, intrusion detection, and cyberattack identification. Current research results confirm the effectiveness and quality of the proposed solution. Experimental evaluations conducted so far demonstrate that fuzzy neural networks can achieve data prediction performance comparable to traditional neural networks. However, the key advantage of fuzzy networks is their significantly lower computational power and RAM requirements. This was achieved through the implementation of ordered fuzzy numbers in the computational processes performed by the artificial neural network. This approach increases the amount of information stored within neurons while maintaining very simple arithmetic operations. Furthermore, the studies revealed a significant increase in training speed compared to existing frameworks. As a result, the solution can be effectively deployed on edge devices and IoT systems, which typically have substantially lower computational capabilities than data centers. The proposed algorithms enable local computation directly on edge devices, eliminating the need to transmit data to centralized computing centers, thereby reducing latency, data transmission costs, and energy consumption. Additionally, the implementation of fuzzy numbers in convolutional neural networks enabled image recognition performance comparable to traditional networks, while requiring less complex network architectures.