EMBER2024: 3.2 Million Malicious and Benign Files Across 6 Formats

3.2 million malicious and benign files across six formats (Win32, Win64, .NET, APK, ELF, PDF) were compiled by joyce8 and released in 2024. This dataset updates previous EMBER iterations to support malware analysis tasks including family classification and behavior prediction. It provides seven distinct label types for multi-faceted security research.

Use Cases

• Binary classification of files using malicious/benign labels
• Malware family classification using family labels
• Malware behavior prediction using behavioral tags

Strengths

• 3.2 million records
• 6 distinct file formats including mobile and document types
• 7 types of labels and tags

Limitations

• Does not include original binary files
• Feature extraction methodology is external to the dataset files

Provenance

Source

joyce8 (Hugging Face) and Arxiv paper 2506.05074

Freshness

Last updated August 2025.