Predicting Dilution in Underground Mines with Stacking Artificial Intelligence Models and Genetic Algorithms

Strategic Insights -

Researchers from Federal University Pernambuco have made significant advancements in predicting dilution in underground mining operations using artificial intelligence. The study highlights the importance of mitigating unplanned dilution caused by blasting inefficiencies or poor rock stability. The research team introduced a statistically rigorous methodology incorporating machine learning algorithms and genetic algorithms to optimize model performance. The results showed that the GA-ANN model outperformed other approaches, achieving high accuracy and low error rates.

Key Takeaways:

  • The researchers developed a statistically rigorous methodology for predicting dilution in underground mining operations using machine learning algorithms and genetic algorithms.
  • The proposed framework incorporates a 10-fold cross-validation procedure with 30 repetitions and nonparametric statistical tests to validate model performance.
  • A total of eight supervised machine learning algorithms were investigated, with their hyperparameters systematically optimized using two distinct genetic algorithm (GA) strategies evaluated under varying population sizes.
  • The models include support vector machines, neural networks, and tree-based approaches, with the GA-ANN model achieving the best performance.
  • The results indicate that the GA-ANN model outperforms other approaches, achieving M A E , R 2 , and R M S E values of 0.2986, 0.8457, and 0.3928 for the training dataset, and 0.1882, 0.9508, and 0.2283 for the testing dataset, respectively.
  • Four stacking models were constructed by aggregating the top-performing base learners, giving rise to ensemble metamodels applied, for the first time, to the task of dilution prediction in underground mining.
  • The research was conducted by Jorge L. V. Mariz and his team, including Tertius S. G. Ferraz, Marinesio P. Lima, Ricardo M. A. Silva, and Hyongdoo Jang, under the guidance of the Computer Center at Federal University Pernambuco.

Statistics:

  • The dataset used for training and testing consisted of 120 samples.
  • The GA-ANN model achieved M A E , R 2 , and R M S E values of 0.2986, 0.8457, and 0.3928 for the training dataset.
  • The GA-ANN model achieved M A E , R 2 , and R M S E values of 0.1882, 0.9508, and 0.2283 for the testing dataset.

Sources:

  • Predicting Dilution in Underground Mines with Stacking Artificial Intelligence Models and Genetic Algorithms (Applied Sciences, 2025,15(11):5996)
  • Applied Sciences (http://www.mdpi.com/journal/applsci)
  • MDPI AG (publisher)
  • Jorge L. V. Mariz (corresponding author)
  • Tertius S. G. Ferraz, Marinesio P. Lima, Ricardo M. A. Silva, Hyongdoo Jang (co-authors)
  • Federal University Pernambuco (research institution)