Authors
1
Geofirst Pty Ltd., 2/7 Luso Drive, Unanderra, NSW 2526, Australia
2
Assistant Professor, Department of Civil Engineering, Tafresh University, Tafresh, Iran
3
Assistant Professor, Department of Civil Engineering, Technical and Vocational University (TVU), Tehran, Iran
4
Assistant Professor, Department of Engineering, Payame Noor University, Tehran, Iran
Abstract
Rock engineering tasks including tunneling, dam building, and ensuring rock slope stability rely heavily on the uniaxial compressive strength (UCS) as a key geomechanical metric. The primary goal of this research was to compare the accuracy of the random forest (RF), k-nearest neighbors (kNN), the decision tree (DT), and the Adaboost in predicting the various rock UCS samples. The approaches were applied to 170 data sets, including point load index (Is(50)), porosity (n), Schmidt hammer (SH), and P-wave velocity (Vp). Initially, the 4 outlier data techniqes were implemented to improve the effectiveness of the used approaches. Then, using the selected data, 4 different machine learning models were developed to predict UCS. Based on different criteria, the 4 models were compared with each other, among which the Adaboost model provided the best response. This model provided R2 values of 0.9631, RMSE of 9.781 and an MAE of 3.684 for the training part and R2 values of 0.9326, RMSE of 13.234 and an MAE of 9.656 for the testing part. Finally, two parameters porosity (n) and Schmidt hammer (SH) were introduced as the most influential parameters in these models.
Keywords
Main Subjects
)