Rock Mechanics
Festus Kunkyin-Saadaari; Jude Baah Offei; Sadique Ibn Sadique; Victor Kwaku Agadzie; Ishamel Abeiku Forson
Abstract
The underground mining operations at the Obuasi Gold Mine rely heavily on the stability of hard rock pillars for safety and productivity. The traditional empirical and numerical methods for predicting pillar stability have limitations, prompting the exploration of advanced machine learning techniques. ...
Read More
The underground mining operations at the Obuasi Gold Mine rely heavily on the stability of hard rock pillars for safety and productivity. The traditional empirical and numerical methods for predicting pillar stability have limitations, prompting the exploration of advanced machine learning techniques. Hence, this work investigates the applicability of stacked generalisation techniques for predicting the stability status of hard rock pillars in underground mines. Four stacked models were developed, using Gradient Boosting Decision Trees (GBDTs), Random Forest (RF), Extra Trees (ET), and Light Gradient Boosting Machines (LightGBMs), with each model taking turns as the meta-learner, while the remaining three models acted as the base learners in each case. The models were trained and tested on a dataset of 201 pillar cases from the AngloGold Ashanti Obuasi Mine in Ghana. Model performance was evaluated using classification metrics, including accuracy, precision, recall, F1-score and Matthews Correlation Coefficient (MCC). The RF-stacked model demonstrated the best overall performance, achieving an accuracy of 93.44%, precision of 94.27%, recall of 93.44%, F1-score of 93.59%, and MCC of 88.90%. Feature importance analysis revealed pillar depth and pillar stress as the most influential factors affecting pillar stability prediction. The results indicate that stacked generalisation techniques, particularly the RF-stacked model, offer promising capabilities for predicting hard rock pillar stability in underground mining operations.
