Introducing stacking machine learning approaches for the prediction of rock deformation. (May 2022)

• Record Type: Journal Article
• Authors: Koopialipoor, Mohammadreza; Asteris, Panagiotis G.; Salih Mohammed, Ahmed; Alexakis, Dimitrios E.; Mamou, Anna; Armaghani, Danial Jahed
• Is Part Of: Transportation geotechnics. Volume 34(2022)
• Publication Date: 2022-05
• Subjects: Granitic rock deformation -- Stacking predictive model -- MPL -- KNN -- RF; Engineering geology -- Periodicals; Soil mechanics -- Periodicals; Rock mechanics -- Periodicals; Transportation -- Periodicals; 624.15105
• DOI: 10.1016/j.trgeo.2022.100756
• Languages: English
• ISSNs: 2214-3912

Abstract

Abstract: Accurate and reliable predictions of rock deformations are crucial in many rock-based projects in civil and mining engineering. In this research, a new system for the prediction of rock deformation was developed using various machine learning models, including multi-layer perceptron (MLP), the k-nearest neighbors (KNN), random forest (RF), and tree. The optimum model developed in this research was designed using a stacking-tree-RF-KNN-MLP structure. The developed structure consolidates different characteristics of four different models with the aim of increasing the prediction accuracy of the Young's modulus. Each of the basic models has various influential parameters that affect the performance of the final system. By optimizing each of these parameters, the stacking-tree-RF-KNN-MLP system was refined to obtain the final model. In this research rock deformations were predicted using four index tests, including porosity, point load strength, Schmidt hammer and p-wave velocity. The stack-tree-KNN-RF-MLP model developed in this research, registered the highest prediction accuracy (R 2 = 0.8197, MSE = 227.371, RMSE = 15.079 and MAE = 12.123). The developed model may be refined over an extended database.