Multivariate Outlier Detection in Geochemical Datasets

Tyler Howe, February 2025

This repository provides an open-source resource for exploring, analyzing, and comparing three different methods of multivariate outlier detection in geochemical datasets in the context of mineral exploration.

Outlier Detection Methods

The three primary outlier detection algorithms explored in this project are:

• Isolation Forest (IF) – (Liu et al., 2008)
• Local Outlier Factor (LOF) – (Breunig et al., 2000)
• Angle-Based Outlier Detection (ABOD) – (Kriegel et al., 2008)

These techniques aim to identify geochemical anomalies that may indicate mineralization zones or other significant geological features.

Background

This work is inspired by Antoine Caté's research on multivariate outlier detection for mineral exploration and builds upon various established anomaly detection methodologies.

Installation & Usage

To run the notebook, ensure you have the following dependencies installed:

Built-in Modules

• time
• importlib
• typing.Tuple

Third-Party Libraries

• pandas
• numpy
• matplotlib.pyplot
• seaborn
• scipy.spatial (cKDTree, KDTree)
• scipy.stats (f_oneway)
• sklearn.ensemble (IsolationForest)
• sklearn.neighbors (LocalOutlierFactor)
• sklearn.preprocessing (StandardScaler)
• sklearn.decomposition (PCA)
• sklearn.metrics (roc_auc_score, roc_curve, auc, mutual_info_score)

References

• Breunig, M.M., Kriegel, H.-P., Ng, R.T., and Sander, J., 2000, LOF: Identifying density-based local outliers: ACM SIGMOD Record, v. 29, no. 2, p. 93–104. https://doi.org/10.1145/335191.335388
• Caté, A., 2025, 6: Multivariate outlier detection for mineral exploration: LinkedIn Pulse, accessed February, 2025, at https://www.linkedin.com/pulse/6-multivariate-outlier-detection-mineral-exploration-antoine-cat%C3%A9-vd4kc/.
• Granitto, M., Schmidt, J.M., Shew, N.B., Gamble, B.M., and Labay, K.A., 2013, Alaska Geochemical Database Version 2.0 (AGDB2)—Including "Best Value" data compilations for geochemical data for rock, sediment, soil, mineral, and concentrate sample media: U.S. Geological Survey Data Series 759, Version 1.0, Denver, CO, accessed February, 2025, at https://doi.org/10.3133/ds759.
• Kriegel, H.-P., Schubert, M., and Zimek, A., 2008, Angle-based outlier detection in high-dimensional data: Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, Las Vegas, Nevada, USA, Association for Computing Machinery, p. 444–452. https://doi.org/10.1145/1401890.1401946.
• Liu, F.T., Ting, K.M., and Zhou, Z.-H., 2008, Isolation Forest: Eighth IEEE International Conference on Data Mining, p. 413–422. https://doi.org/10.1109/ICDM.2008.17.
• Maklin, C., 2022, Isolation Forest: Cory Maklin - Medium, accessed at https://medium.com/@corymaklin/isolation-forest-799fceacdda4.
• Saudi Geological Survey (SGS), [2023], [Dataset RGP GSAS Geochemical survey Jabal Al Hasir]: Saudi Geological Survey Data Portal, accessed [February, 2025], at [https://ngdp.sgs.gov.sa/ngp/].
• Saudi Geological Survey (SGS), [2023], [Mineral Occurrence Documentation System (MODS) Database]: Saudi Geological Survey Data Portal, accessed [February, 2025], at [https://ngdp.sgs.gov.sa/ngp/].
• Shahrestani, S., and Sanislav, I., 2025, Mapping geochemical anomalies using angle-based outlier detection approach: Journal of Geochemical Exploration, v. 269. https://doi.org/10.1016/j.gexplo.2024.107635
• U.S. Geological Survey (USGS), 2008, Alaska Resource Data File (ARDF): U.S. Geological Survey Open-File Report 2008-1225, accessed February, 2025, at https://mrdata.usgs.gov/ardf/.