Original Dataset for Paper: Source Identification of Heavy-Metal Hazardous Solid Wastes Using Robust Mineral-Phase Fingerprints

Improper disposal of heavy-metal hazardous solid waste (HMHSW) poses serious environmental and public-health risks. Reliable source identification, critical for enforcing accountability and guiding remediation, remains challenging due to the limitations of conventional chemical-fingerprinting approaches, which are vulnerable to process-related fluctuations and waste mixing. Herein, we develop a novel and noise-robust mineral-phase fingerprinting method inspired by the typomorphic mineral assemblage concept in genetic mineralogy. Using a comprehensive dataset comprising 159 waste codes from 25 sources, we designed a knowledge-guided noise-injection strategy that simulates seven representative noisy scenarios. From these, a robust fingerprint set of 153 mineral phases was selected based on weighted stability scores. This strategy markedly improved the machine-learning models’ performance and substantially reduced overfitting relative to baseline models. The optimal model achieved average balanced accuracy, macro F1, and macro AUC values exceeding 0.90 under noisy test conditions, while providing reliable uncertainty estimates. SHAP analysis confirmed that the model’s predictions are driven by source-specific typomorphic mineral phases. Overall, this work establishes a robust and interpretable forensic tool for HMHSW source tracing and demonstrates the strong potential of mineral phases as diagnostic tracers for waste tracking.