Abstract: We propose a universal method for threat detection with a set-valued evidential classification based on Dempster-Shafer theory and deep learning. Our approach is designed to handle the ...