Integrated ore classification using stand-alone and hybridised machine learning algorithms.

Area of Science:

• Geosciences and Mining
• Artificial Intelligence and Machine Learning

Background:

• Accurate estimation of gold ore grades is crucial for efficient mining operations.
• Traditional methods can be labor-intensive and may lack precision in complex geological settings.
• Trace element analysis provides valuable geochemical signatures for ore body characterization.

Purpose of the Study:

• To develop a robust artificial intelligence (AI) classification model for estimating gold ore grades.
• To integrate various machine learning (ML) algorithms within a committee machine (CM) framework for enhanced performance.
• To compare the effectiveness of stand-alone ML algorithms against optimized committee machine models.

Main Methods:

• Utilized inductively coupled plasma (ICP) analysis of 19 trace elements from eight drill holes as input features.
• Implemented eight stand-alone ML classifiers (e.g., AdaBoost, XGBoost, ANFIS) for initial classification.
• Employed genetic (GA) and simulated annealing (SA) optimization algorithms within a CM framework to create a unified, optimized model (CMSA).

Main Results:

• The AdaBoost algorithm showed superior performance among the stand-alone ML classifiers.
• The committee machine with simulated annealing (CMSA) model outperformed the GA optimizer and stand-alone algorithms.
• CMSA achieved a 7.28% improvement in overall classification accuracy compared to the average performance of stand-alone algorithms.

Conclusions:

• AI-driven classification models, particularly ensemble methods like CMSA, offer a powerful approach for accurate gold ore grade estimation.
• Trace element data, when processed through sophisticated ML algorithms, can effectively delineate ore, low-grade ore, and waste classes.
• The developed CMSA model provides a robust and accurate tool for geological studies and mineral exploration.