Qmin

Qmin automates processing and analysis of mineral chemistry datasets from electron probe micro-analyzers (EPMA) to classify minerals, impute missing values, and predict mineral formulas for mineralogical research and prospecting.

Key Features:

• Mineral Classification: Employs a hierarchical structure of classifiers to categorize minerals based on their chemical compositions.
• Missing Value Imputation: Uses multivariate regression techniques to fill gaps in mineral chemistry datasets.
• Mineral Formula Prediction: Predicts mineral formulas with nested models built from Random Forest classification and regression algorithms.

Scientific Applications:

• Mineralogy and Prospecting: Applies to mineralogical studies and prospecting workflows, including diamond prospecting, using EPMA-derived compositions and a training database from the GEOROC repository.
• Validation and Performance: Demonstrated performance in a blind test of over 11,000 analyses from the Diamante Brasil Project with a balanced classifier accuracy of approximately 99% for known minerals.

Methodology:

Uses Random Forest algorithms with a hierarchical structure of classifiers and regressors, nested Random Forest classification and regression models for formula prediction, and multivariate regression for missing value imputation.