Abstract:
Underground mine tunnel failures pose severe safety and operational challenges, particularly in
structurally complex regions such as the Greenstone Belt of Tiira, Uganda. These failures are
commonly driven by poor geotechnical design, unassessed rock properties, and excessive
overburden stress, often resulting in injuries, fatalities, equipment loss, and halted operations. To
address this, the study aimed to develop a predictive model to assess tunnel failure risks and
improve the design of underground mines.
Field and laboratory data were collected from both failed and stable tunnel zones, focusing on
parameters such as rock stress, specific gravity, tunnel dimensions, and overburden pressure.
Laboratory analysis included uniaxial compressive strength testing. Statistical tools including
Kendall’s Tau and Spearman’s rank correlation were employed to identify key predictors. A
machine learning model, specifically a Random Forest Classifier (RFC), was trained on the
labeled dataset and evaluated using performance metrics like accuracy, precision, recall, and
AUC. The RFC model achieved high predictive performance: 90.5% accuracy, 92.3% precision, and an
AUC of 0.99. Sensitivity analysis revealed rock strength, specific gravity, and overburden stress
as the most influential variables affecting tunnel stability. Conversely, wider tunnel dimensions
were associated with increased failure risk.
The study demonstrates the potential of integrating machine learning into geotechnical
engineering to predict tunnel failures effectively. This approach supports data-driven design in
early-stage mine planning and enhances operational safety. It also contributes to sustainable
mining practices aligned with SDG 8 and SDG 9.
