CLAMPD-1905-Dataset

CLAMPD-1905 is the first comprehensive cross-language dataset for malicious package detection research, offering unified feature representations and full reproducibility across PyPI and npm ecosystems. This dataset enables standardized benchmarking of malware detection tools across diverse software supply chain ecosystems. Built on curated packages from MalwareBench, CLAMPD-1905 includes 23,764 balanced software packages (11,882 benign and 11,882 malicious) from PyPI and npm, each represented as a 1905-dimensional multimodal feature vector. These vectors combine static metadata (9 features), dependency graph embeddings (128 dimensions), and API behavioral features (1768 dimensions) extracted from over one million source files. This unified representation eliminates ecosystem-specific preprocessing requirements, facilitates reproducible experiments, and supports research on cross-ecosystem supply chain security threats. Key Features Cross-Ecosystem Coverage: First dataset to provide unified feature representation across PyPI and npm Multimodal Feature Integration: Combines normalized metadata, HAN dependency graph embeddings, and BERT API behavioral features Perfect Class Balance: 23,764 packages with equal distribution of benign and malicious samples (11,882 each) Advanced Feature Engineering: 1905-dimensional vectors through sophisticated multimodal fusion Research Ready: Standardized format for immediate model training and evaluation Reproducible: Complete feature engineering pipeline and pre-computed embeddings Comprehensive Representation: Captures structural, behavioral, and contextual package characteristics