Machine Learning in Surface Mining Exploitation – A Systematic Review

The extractive industry is crucial for the global supply of raw materials but is facing in creasing complexity due to the scarcity of shallow deposits and increasingly challenging geological conditions. Machine Learning (ML) has started to being utilized as a promising tool to optimize unitarian operations and enhance decision-making through high predictive models. The present systematic review follows the PRISMA guidelines, including 57 articles published between 2020 and 2025 from four databases. The methodological rigor was ensured by utilizing the ROBINS-I tool for bias assessment. Studies were categorized into four categories: blasting phase, load and haul, extraction and overall exploitation. Results indicate that articles are highly concentrated in the blasting phase (75% of the articles), with hybrid architectures like XGBoost and Support Vector Machine achieving high reliability with R2 exceeding 0.97. In the other hand, sectors such as Load and Haul tend to rely on Reinforcement Learning and Computer Vision to address dynamic operational complexities. Although these advancements have been made, the transition to Mining 4.0 is being restrained by site-specific characteristics of the datasets and a high lack of transparency. Future research must prioritize standardized reporting and the inclusion of operational efficiency indicators to ensure the scalable integration of ML into real-word mining environments.