StaticMalFiles: A Heterogeneous Dataset for Static Malicious File Detection and Machine Learning

Abstract

StaticMalFiles is a heterogeneous dataset developed for research on static malicious file detection using machine learning techniques. Unlike benchmark datasets distributed with pre-extracted attributes, this dataset was constructed from raw benign and malicious files collected from publicly available repositories and software distribution platforms. A custom feature extraction framework was used to derive structural, metadata-based, content-based, and instruction-level characteristics directly from binary artifacts. The dataset contains: 1,237 labeled samples 703 benign files 534 malicious files 30 extracted features Multiple file formats including executables, libraries, scripts, installers, archives, and Unix executables The extracted features cover four major categories: Metadata features (extension, architecture, encoding, digital signature) Structural features (entropy, sections, packers, resources) Content-based features (Unicode strings, suspicious strings, suspicious functions) Instruction-level features (opcode statistics, instruction counts, instruction diversity) The dataset was created as part of a research effort supporting reproducible machine learning experiments for malware detection and explainable cybersecurity analytics. Besides the dataset itself, the repository also provides the feature extraction tools, preprocessing pipeline, and documentation required to reproduce the complete workflow. Potential applications include: Malware detection Malicious file classification Feature engineering research Explainable AI (XAI) Cyber threat intelligence Static malware analysis Benchmarking machine learning models Researchers are encouraged to use, extend, and compare their methods against this dataset to promote reproducibility and transparency in cybersecurity research.