SOREL-20M Subset: 196,534 Malware Samples with EMBER v2 Features

SOREL-20M Subset Dataset contains 196,534 samples for malware detection, created by reveng-grp-2025. It includes 99,506 malicious and 97,028 benign samples, each described by 2,351 EMBER v2 features. The dataset was last updated on Hugging Face on 2025-06-02.

Use Cases

• Training binary classifiers for malware detection based on EMBER v2 features.
• Benchmarking feature extraction and model performance on a large-scale malware dataset.
• Analyzing the distribution and characteristics of malicious versus benign software samples.

Strengths

• Contains 196,534 total samples, providing a substantial dataset for model training.
• Includes a balanced split with 99,506 malicious and 97,028 benign samples.
• Features are derived from the EMBER v2 feature set, a standard in static malware analysis.

Limitations

• Row count and file size are unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

reveng-grp-2025 on Hugging Face.

Collection Method

Subset of the larger SOREL-20M dataset.

Freshness

Last updated 2025-06-02 11:38:12.