Application of the CIPW Normative Method Using Artificial Intelligence

Integration of classical CIPW normative petrology with machine learning for igneous rock classification.

Overview

This project integrates classical CIPW normative petrology with modern machine learning techniques for the systematic classification of igneous rocks.

Methodology

CIPW normative calculations following the Cross–Iddings–Pirsson–Washington formulation
Compilation and QA/QC of major-oxide geochemical datasets
Feature extraction from normative mineral assemblages
Training of Random Forest, XGBoost, and SVM models
Dimensionality reduction using PCA
Stratified cross-validation

Dataset

Initial compilation: 3,358 igneous rock samples from Peru
Final validated dataset: 2,707 samples after QA/QC and standardization

Results

Volcanic rocks: classification accuracy ≈ 0.80–0.82
Plutonic rocks: classification accuracy ≈ 0.52–0.56

Performance differences reflect mineralogical complexity and compositional overlap.

Contributions

The project demonstrates the integration of petrological reasoning with data-driven classification approaches.