Application of soft computing methods in underground mine stability: A PRISMA-based systematic review

Geological uncertainties make the prediction of rock mass behavior highly challenging, emphasizing the need for continuous advances in excavation stability and risk mitigation. Soft computing algorithms, particularly machine learning (ML) and metaheuristics, have gained prominence as complementary tools in this domain. However, systematic reviews of their applications in underground mining remain limited. This study conducts a comprehensive review of ML and metaheuristic algorithms for excavation stability analysis based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework from six major scientific databases over a 25-year period. From 1379 retrieved records, 239 articles were analyzed, revealing five thematic clusters: Geohazards (48%), Underground Stability (25%), Rock Mass Characterization (15%), Ground Support (8%), and Blasting (4%). Research activity has expanded substantially since 2016, with geohazard-related studies accelerating after 2021. ML dominates the field (71%), while metaheuristics (2%) and hybrid approaches (27%) are smaller but expanding contributors. Genetic Algorithms and Particle Swarm Optimization are the most common metaheuristics, while ML is led by Multilayer Perceptrons, Support Vector Machines, Random Forest, and XGBoost. Hybrid approaches consistently achieved superior predictive accuracy compared to ensemble or single ML models. Overall, the findings highlight hybridized soft computing as a promising direction for advancing reliable, data-driven stability assessments in complex underground mining environments. Nonetheless, ML in underground rock mechanics is constrained by geological complexity, data scarcity, and limited interpretability, but these challenges can be mitigated through physics-informed, uncertainty-aware, and hybrid ML–rock mechanics frameworks.