BODMAS: An Open Dataset for Learning based Temporal Analysis of PE Malware
Limin Yang, Arridhana Ciptadi, Ihar Laziuk, Ali Ahmadzadeh, Gang Wang
Abstract
We describe and release an open PE malware dataset called BODMAS to facilitate research efforts in machine learning based malware analysis. By closely examining existing open PE malware datasets, we identified two missing capabilities (i.e., recent/timestamped malware samples, and well-curated family information), which have limited researchers’ ability to study pressing issues such as concept drift and malware family evolution. For these reasons, we release a new dataset to fill in the gaps. The BODMAS dataset contains 57,293 malware samples and 77,142 benign samples collected from August 2019 to September 2020, with carefully curated family information (581 families). We also perform a preliminary analysis to illustrate the impact of concept drift and discuss how this dataset can help to facilitate existing and future research efforts.