Evolutionary Random Forest Algorithms for Predicting the Maximum Failure Depth of Open Stope Hangingwalls

The prediction of the failure depth of open stope hangingwalls (HWs) has become a crucial task for underground mines worldwide. In this paper, an evolutionary random forest (RF) technique that combines the RF algorithm and particle swarm optimization (PSO) was proposed to model the non-linear relationship between maximum failure depth (MFD) of stope HWs and its influencing variables. The training and verification of RF models were conducted on a dataset collected from the literature, and a total number of 125 valid HW cases were analyzed. The 13 influencing variables were used as inputs. Hyper-parameters of RF models were tuned by PSO with fivefold cross-validation, and the performance of RF models was measured by root-mean-square error and correlation coefficient (R). Experimental results show that the RF algorithm had great potential for predicting the MFD of HWs. The R values between the predicted and actual MFD values were 0.93 and 0.84, respectively. The stope design method was found to be the most significant influencing variable with an importance score of 0.516 out of 1, followed by the RQD of rock mass (0.127), stope strike (0.078), and HW height (0.075).