An Adaptive Deep Learning-based ARP Spoofing Detection Framework with Hybrid Thresholding

Address Resolution Protocol (ARP) spoofing is a significant cybersecurity threat that allows attackers to manipulate network traffic, resulting in data interception and service disruptions. A deep learning-based ARP spoofing detection framework is introduced, incorporating a Temporal Convolutional Network (TCN) alongside a combination of static and dynamic thresholding methods. Specifically, an Exponential Moving Average (EMA)-based dynamic thresholding technique is utilized to dynamically identify anomalies. To address class imbalance challenges, the SMOTE is employed, enhancing classification accuracy. Experimental results show 99.53% accuracy, along with excellent precision, recall, and F1 Score. The proposed Adaptive Thresholding- based Deep Learning model (ATDL) efficiently reduces false positives while maintaining superior detection performance.