An ensemble-based strategy for robust predictive volcanic rock typing efficiency on a global-scale: A novel workflow...

Area of Science:

• Geochemistry
• Petrology
• Data Science

Background:

• Traditional rock classification relies on laboratory measurements, paleontological data, and well-logs.
• Digital intelligence offers potential for enhanced accuracy and speed in rock classification.

Purpose of the Study:

• To develop a comprehensive approach for categorizing igneous rock types using global geochemical data.
• To integrate advanced clustering, classification, data mining, and statistical methods.

Main Methods:

• Utilized a global geochemical dataset of ~25,000 points from 15 igneous rock types.
• Employed hierarchical clustering, linear projection analysis, and multidimensional scaling.
• Evaluated eight classifiers, including Logistic Regression, Gradient Boosting, Random Forest, K-nearest Neighbors, Support Vector Machine, Artificial Neural Network, and two ensemble models (EN-1, EN-2).

Main Results:

• The ensemble model EN-2 achieved the highest accuracy at 99.2%, outperforming individual models and EN-1 (98%).
• Artificial Neural Network (ANN) achieved 98.2% accuracy.
• Silicon dioxide (SiO2), potassium oxide (K2O), and sodium oxide (Na2O) were identified as the most important features for classification.

Conclusions:

• Ensemble models significantly enhance the accuracy and reliability of igneous rock classification.
• The developed approach provides more precise global rock typing by capturing complex data patterns.
• Machine learning techniques offer a powerful tool for advancing geological sample analysis.