Application of Distance Metric Learning to Automated Malware Detection

Distance metric learning aims to find the most appropriate distance parameters improve similarity-based models such as k-Nearest Neighbors or k-Means. In this paper, we apply problem of malware detection. We focus on two tasks: (1) classify and benign files with a minimal error rate, (2) detect much possible while maintaining low false positive rate. propose detection system using Particle Swarm Optimization that finds feature weights optimize similarity measure. compare performance approach three state-of-the-art techniques. metrics trained in way lead significant improvements classification. conducted evaluated experiments more than 150,000 Windows-based samples. Features consisted metadata contained headers executable portable file format. Our experimental results show our based achieves 1.09 % rate at 0.74 (FPR) outperforms all machine algorithms considered experiment. Considering second task related keeping FPR, achieved 1.15 only 0.13 FPR.