Detecting and responding to attacks and weather effects in hybrid FSO/RF systems using the Dempster-Schaffer theory with AI algorithms

Abstract

In this paper, a new intelligent switch for hybrid
Free-Space Optical (FSO) RF communication is proposed for
improved reliability and security in the presence of dynamic
environmental changes and cyber-attack interferences. Using
Dempster-Shafer Theory (DST) for reliable threat classification
and ANN, KNN, and SVM for machine learning, an extraordinary
real-time communication link selection is achieved. A
broad training dataset (10,000 simulated samples), covering
eavesdropping and jamming threats, fog and dust effects, was
used to train and validate the network. Our work combines
DST to combine evidence from multiple sources and make an
accurate belief assignment for communication modes. In addition,
the system exhibits a high claimed confidence, RF and FSO link
beliefs around 0.88-0.89 and 0.82-0.83, respectively. The machine
learning models have excellent performance on threat detection
and mode classification. ANN, KNN, and SVM obtained accuracies
of 0.99986, 0.99984, and 0.99930, respectively. All models
achieved near-perfect AUC values, where most classes reach 1,
meaning a better discriminative performance. Importantly, the
performance of ANN was significantly outperformed by KNN
and SVM in all metrics, demonstrating its robustness. This
work provides an efficient and dynamic approach to keep the
communication in difficult FSO/RF links secure and reliable,
and brightens the path for future communication systems.