Paper contribution

Overbreak/underbreak prediction in drill-and-blast tunnelling using machine learning: automated feature selection from MWD data

Alla Sapronova; Thomas Marcher

Proceedings
EUROCK 2026
Pages
1377-1382
Submitted
28 January 2026
Accepted
15 June 2026
Published
12 August 2026

Summary

Abstract

In this work, an automatic data-driven feature selection strategy was applied to select the subset of input parameters that govern the outcome in the drill-and-blast tunnelling method. The features include geometric drilling variables and measurement-while-drilling (MWD) data, which capture the mechanical response of the rock mass during drilling. The objective was to determine a subset of variables with the highest informational content with respect to the target variable: a deviation from the design excavation profile (overbreak/underbreak). By limiting the training data to a set of highly informative inputs with minimised redundancy, this approach aims to enhance the generalisation and robustness of the machine learning model (MLM). The selection process combined statistical, exploratory, and model-based analyses by computing, for each variable, (i) a redundancy score from correlation analysis, (ii) a variance-structure contribution score from a principal component analysis (PCA) loading–based metric over the retained principal components, and (iii) a predictive contribution score from mutual information (MI) and feature-importance estimates from Light Gradient Boosting Machine (LGBM), Random Forest Regression (RFR), and Kolmogorov–Arnold Networks (KANs). The individual, per-variable scores were then normalised and aggregated by averaging to produce a final consensus ranking for each variable. This multi-criteria approach enabled the removal of weak or redundant predictors. The results demonstrate that the relative importance of each variable is primarily defined by the dataset’s internal statistical relationships and interdependencies. Features, established in the domain knowledge as critical for predicting tunnelling outcomes, may be redundant or exhibit insufficient variance, thereby not contributing to prediction accuracy. Specifically, the spacing between neighbouring contour holes was found to contribute little to MLM outcomes. Also, it is shown that selecting input features is more important than the algorithmic complexity of the MLM: simpler models can achieve higher accuracy when trained on informative input variables.

Contributors

Authors

  1. Alla Sapronova Graz University of Technology, Austria ORCID 0000-0003-2863-1536
  2. Thomas Marcher Graz University of Technology, Austria

Index terms

Keywords

Full text

Attachments

Works cited

References

  1. Im H, Kurauchi T, Sato N, Kawamura Y, Jang H, Topal E (2025) Tunnel overbreak prediction: an integrated approach using 3D photogrammetry and machine learning. Mining, Metallurgy & Exploration 42:1441–1457. https://doi.org/10.1007/s42461-025-01278-1
  2. Liu Y, Li A, Zhang H, Wang J, Li F, Chen R, Wang S, Yao J (2023) Minimization of overbreak in different tunnel sections through predictive modeling and optimization of blasting parameters. Frontiers in Ecology and Evolution 11:1255384. https://doi.org/10.3389/fevo.2023.1255384
  3. Liu Z, Wang Y, Vaidya S, Ruehle F, Halverson J, Soljačić M, Hou TY, Tegmark M (2025) KAN: Kolmogorov-Arnold Networks. Proceedings of the 13th International Conference on Learning Representations, ICLR 2025. OpenReview. https://openreview.net/forum?id=Ozo7qJ5vZi
  4. Navarro J, Sanchidrián JA, Segarra P, Castedo R, Costamagna E, López LM (2018) Detection of potential overbreak zones in tunnel blasting from MWD data. Tunnelling and Underground Space Technology 82:504–516. https://doi.org/10.1016/j.tust.2018.08.060
  5. Qiu Y, Zhou J, He B, Armaghani DJ, Huang S, He X (2024) Evaluation and interpretation of blasting-induced tunnel overbreak: using heuristic-based ensemble learning and gene expression programming techniques. Rock Mechanics and Rock Engineering 57:7535–7563. https://doi.org/10.1007/s00603-024-03947-x
  6. Sakai K, Miyanaga S, Yamagami M, Sapronova AV, Soliman A (2026) Study on machine learning algorithm for optimal tunnel blasting design. In: Deb D et al. (eds) Advances in Rock Mechanics—Infrastructure Development, Volume 1. Lecture Notes in Civil Engineering 783, Springer, Singapore, pp 245–254. https://doi.org/10.1007/978-981-95-4259-8_26
  7. Sapronova A, Marcher T (2025) Improving data quality with advanced pre-processing of MWD data. Geotechnics 5(2):28. https://doi.org/10.3390/geotechnics5020028
  8. Schunnesson H (1996) RQD predictions based on drill performance parameters. Tunnelling and Underground Space Technology 11(3):345–351. https://doi.org/10.1016/0886-7798(96)00024-7
  9. Schunnesson H (1998) Rock characterisation using percussive drilling. International Journal of Rock Mechanics and Mining Sciences 35(6):711–725. https://doi.org/10.1016/S0148-9062(97)00332-X
  10. Teale R (1965) The concept of specific energy in rock drilling. International Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts 2(1):57–73. https://doi.org/10.1016/0148-9062(65)90022-7
  11. van Eldert J, Schunnesson H, Johansson D, Saiang D (2020) Application of measurement while drilling technology to predict rock mass quality and rock support for tunnelling. Rock Mechanics and Rock Engineering 53:1349–1358. https://doi.org/10.1007/s00603-019-01979-2
  12. van Eldert J, Schunnesson H, Saiang D, Funehag J (2020) Improved filtering and normalizing of measurement-while-drilling data in tunnel excavation. Tunnelling and Underground Space Technology 103:103467. https://doi.org/10.1016/j.tust.2020.103467
  13. Zhang Y, Zhou J, Li J, He B, Armaghani DJ, Huang S (2025) Advancing overbreak prediction in drilling and blasting tunnel using MVO, SSA and HHO-based SVM models with interpretability analysis. Geomechanics and Geophysics for Geo-Energy and Geo-Resources 11:53. https://doi.org/10.1007/s40948-025-00963-1