MLRan
收藏资源简介:
MLRan是一个大型的行为性勒索软件数据集,包含超过4800个样本,跨越64个勒索软件家族和一个平衡的良性软件样本集。样本时间跨度从2006年到2024年,涵盖了四大主要类型的勒索软件:锁屏、加密、勒索软件即服务和现代变种。数据集专注于针对Windows系统的勒索软件,因为它涵盖了95%的勒索软件文件。MLRan捕获了九个关键行为特征,包括API调用、注册表键、文件和目录操作、字符串、网络活动、系统进程、丢弃的文件和数字签名。此外,论文提出了构建高质量行为勒索软件数据集的指南(GUIDE-MLRan),并使用这些指南确保了MLRan数据集的严格管理。为了简化恶意软件的动态分析,论文中提到了增强Cuckoo Sandbox的功能,以自动化文件提交和分析结果排序。为了降低数据维度,论文提出了一种基于互信息过滤的特征选择策略,将最初的640万个特征减少到24162个,然后通过递归特征消除得到483个高度信息化的特征。最后,论文使用SHAP和LIME等可解释AI技术分析了关键勒索软件行为,并公开了数据集和工具实现,以支持可重复性和鼓励未来研究。
MLRan is a large-scale behavioral ransomware dataset containing over 4,800 samples spanning 64 ransomware families, alongside a balanced set of benign software samples. The samples cover a time span from 2006 to 2024, and include four major categories of ransomware: screen-locker, encryptor, ransomware-as-a-service (RaaS), and modern variants. This dataset focuses on Windows-targeted ransomware, as it covers 95% of ransomware instances in the wild. MLRan captures nine critical behavioral features, including API calls, registry keys, file and directory operations, strings, network activity, system processes, dropped files, and digital signatures. Additionally, the paper proposes GUIDE-MLRan, a guideline for constructing high-quality behavioral ransomware datasets, and leverages this framework to ensure strict curation of the MLRan dataset. To simplify dynamic malware analysis, the paper describes enhancements to the Cuckoo Sandbox framework for automating sample submission and analysis result ranking. To reduce data dimensionality, the paper presents a mutual information-based feature selection strategy, which reduces the initial 6.4 million features to 24,162, followed by recursive feature elimination (RFE) to obtain 483 highly informative features. Finally, the paper analyzes key ransomware behaviors using explainable AI (XAI) techniques including SHAP and LIME, and releases the dataset and tool implementations to support reproducibility and encourage future research.




