PRISM: PE Relational Inter-Section Matrix — A 2D Section-Aware Dataset for Static PE Malware Detection

PRISM encodes each Windows PE binary as a 2D matrix (17×26 physical; 25 effective features per section row, column 22 reserved/zero) in which rows are PE sections in file order and a global summary row provides an EMBER-compatible file-level descriptor. This deposit contains the PRISM feature matrices (full deduplicated corpus of 83,633 unique matrices and the 49,204-sample family-filtered analysis corpus), labels and metadata, trained LightGBM baseline models (B1-proxy, B2, B3, B4), EMBER feature vectors extracted from restored BODMAS binaries, and reproduced separability/benchmark result files. No malware binaries are distributed. Feature matrices are released under CC BY 4.0; accompanying code (on GitHub) under MIT.