Efficiency of Classification Algorithms for Prediction of Rock Mass ÖNORM B Class in Himalayan Tunnelling

Abstract

For underground excavation, accurate assessment of rock mass behaviour is imperative for a robust design of support system. For the ambitious, broad gauge rail link project in Uttarakhand Himalaya, India, apart from Q, and RMR, ÖNORM B system is being used for tunnel support recommendation. However, ÖNORM B system is qualitative in nature, and thus measurement of surrounding rock mass deformation is measured to designate the rock class. This approach is expensive and not often feasible. Therefore, present study attempts, perhaps for the first time, to quantify the prediction of ÖNORM B class of rock mass, using five easy to assess parameters. Two parameters from RMR, two from Q-system, and one common in both were used as inputs. Nine standard machine learning classifiers have been trained on 873 rows of data, and validated on 218 data points. Accuracy, precision, and ROC were evaluated for each classification algorithm. Results are quite promising with highest accuracy and precision in predicting the ÖNORM B class, delivered by Extra Tree, Random Forest, and Decision Tree classifiers. However, the authors recommend Extra Tree classifier since they are the least prone to overfitting and can be generalized. © The Author(s), under exclusive licence to Indian Geotechnical Society 2025.