Combining supervised and unsupervised learning for zero-day malware detection

摘要: Malware is one of the most damaging security threats facing Internet today. Despite burgeoning literature, accurate detection malware remains an elusive and challenging endeavor due to increasing usage payload encryption sophisticated obfuscation methods. Also, large variety classes coupled with their rapid proliferation polymorphic capabilities imperfections real-world data (noise, missing values, etc) continue hinder use more algorithms. This paper presents a novel machine learning based framework detect known newly emerging at high precision using layer 3 4 network traffic features. The leverages accuracy supervised classification in detecting adaptability unsupervised new classes. It also introduces tree-based feature transformation overcome issues construct informative features for task. We demonstrate effectiveness real from service provider.